AI and policing

The benefits and

challenges of artificial
intelligence for
law enforcement

An Observatory Report from the Europol Innovation Lab

AI AND POLICING
THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT
An Observatory Report from the Europol Innovation Lab

PDF | ISBN 978-92-95236-35-6 | ISSN 2600-5182 | DOI: 10.2813/0321023 | QL-01-24-000-EN-N

Neither the European Union Agency for Law Enforcement Cooperation nor any person acting
on behalf of the agency is responsible for the use that might be made of the following information.

Luxembourg: Publications Office of the European Union, 2024

© European Union Agency for Law Enforcement Cooperation, 2024

Reproduction is authorised provided the source is acknowledged.

For any use or reproduction of photos or other material that is not under the copyright of
the European Union Agency for Law Enforcement Cooperation, permission must be sought
directly from the copyright holders.

While best efforts have been made to trace and acknowledge all copyright holders, Europol
would like to apologise should there have been any errors or omissions. Please do contact us
if you possess any further information relating to the images published or their rights holder.

Cite this publication: Europol (2023), The Second Quantum Revolution – The impact of quantum computing
and quantum technologies on law enforcement, Europol Innovation Lab observatory report,
Publications Office of the European Union, Luxembourg.

This publication and more information on Europol are available on the Internet.
www.europol.europa.eu

3
 Contents

6 Foreword

7 Executive Summary

8 Introduction
8 Background
9 Objectives

10 Key takeaways for law enforcement

12 Applications of AI in law enforcement

12 Data analytics
Large and complex data sets
Predictive Policing
OSINT and SOCMINT
Natural Language Processing (NLP)

20 Digital forensics
21 Computer vision and biometrics
Video monitoring and analysis
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

Image classification
Biometrics
Biometric categorisation

28 Improved resource allocation and strategic planning

29 Generative AI

31 Technological limitations and challenges

32 Ethical and social issues in AI for law enforcement

32 Data bias and fairness
33 Privacy and surveillance
34 Accountability and transparency
36 Human Rights and Discrimination

37 The EU Artificial Intelligence Act: overview and context

37 Objectives, scope and key provisions
Prohibited uses of AI
Law enforcement exceptions to prohibited practices
High-Risk AI Systems
The filter mechanism for the evaluation of high-risk systems

4
43 Implications for law enforcement agencies
46 Innovation and regulatory sandboxes

47 Balancing the benefits and restrictions

47 Addressing concerns of bias and discrimination
48 Safeguarding privacy and data protection

49 Future outlook and recommendations

49 Potential for technological advancements
50 Building public trust and acceptance
51 Strengthening collaboration and knowledge sharing within LEAs

53 Conclusion

54 AI Glossary

56 Endnotes

5
 Foreword Artificial Intelligence (AI) will profoundly alter the landscape of law
enforcement, offering innovative tools and opportunities to enhance
our capabilities in safeguarding public safety. This flourishing
technological field promises to revolutionise how we analyse
complex data sets, improve forensic methodologies, and develop
secure communication channels.

However, alongside these advancements, AI introduces new

challenges and potential vulnerabilities, particularly in areas
like data privacy and the integrity of AI-driven decisions.
It is imperative that we navigate these advancements with
a strategic approach, balancing innovation with the ethical
implications and societal impact.

As Europol stands at the forefront of embracing technological

innovation within law enforcement, we are acutely aware of the
necessity to stay ahead of these developments. This Europol
Innovation Lab Observatory report, serves not only as a testament
to our commitment to embracing AI responsibly, but also as a guide
for the European law enforcement community as we step into this
new era of digital policing.

I hope that this report will contribute to shed light on the intricate
dynamics of AI for policing, providing valuable insights for our
stakeholders and helping the law enforcement community on its
path towards adopting AI’s potential responsibly. Together, we
embark on this journey, ready to face the challenges and seize
the opportunities that the AI revolution presents, ensuring that we
continue to protect and serve our communities in an increasingly
digital world.
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

Catherine De Bolle
Executive Director of Europol

6
Executive This report aims to provide the law enforcement community with
a comprehensive understanding of the various applications and
Summary uses of artificial intelligence (AI) in their daily operations. It seeks
to serve as a textbook for internal security practitioners, offering
guidance on how to responsibly and compliantly implement AI
technologies. In addition to showcasing the potential benefits and
innovative applications of AI, such as AI-driven data analytics, the
report also aims to raise awareness about the potential pitfalls and
ethical considerations of AI use in law enforcement. By addressing
these challenges, the report endeavours to equip law enforcement
professionals with the knowledge necessary to navigate the
complexities of AI, ensuring its effective and ethical deployment
in their work. The report focuses on large and complex data sets,
open-source intelligence (OSINT) and natural language processing
(NLP). It also delves into the realm of digital forensics, computer
vision, biometrics, and touches on the potential of generative AI.

The use of AI by law enforcement is increasingly scrutinised due to

its ethical and societal dimensions. The report attempts to address
concerns about data bias, fairness, and potential encroachments on
privacy, accountability, human rights protection and discrimination.
These concerns become particularly relevant in the context of the
EU’s Artificial Intelligence Act (EU AI Act), an overview of which is
detailed in this report, as well as its broader context. The report
emphasises the significance of the forthcoming regulation, detailing
its objectives, scope, and principal provisions. The Act’s implications
for law enforcement agencies are also discussed, emphasising
the balance between fostering innovation and ensuring ethical use
beyond compliance.

Central to the report is the assessment of how law enforcement can

maintain a delicate balance between leveraging AI’s benefits and
addressing its inherent restrictions. Strategies for addressing bias,
privacy concerns, and the pivotal role of accountability frameworks,
are elaborated. The report highlights the importance of innovative
regulatory environments.

The concluding section forecasts the trajectory of AI in

law enforcement, underscoring the potential technological
advancements on the horizon. It also emphasises the need for
public trust and acceptance, and the importance of collaboration
and knowledge sharing. This comprehensive document serves
as both a guide and a reflective tool for stakeholders vested in
the confluence of AI and law enforcement within the European
landscape.

7
 Introduction Background

In the ever-evolving landscape of law enforcement, artificial

intelligence (AI) has emerged as a transformative tool, bringing
capabilities that could completely reshape policing. Law
Enforcement Agencies (LEAs), both in the European Union
(EU) and globally are confronted with increasingly complex
challenges. From the exponential growth in data generated by
digital devices and online services to the complex nature of
modern criminal activities, it is evident that traditional policing
methods alone are not sufficient as a response. Moreover, the
globalisation of crime1, marked by cyber threats, cross-border
trafficking, and international terrorism, presents an increasingly
challenging landscape that calls for advanced and innovative
solutions.
In light of this, AI offers a promising alternative. By employing
state-of-the-art technologies law enforcement can address
many of these pressing challenges. The power of AI in
processing vast amounts of data, and filtering for relevant
content, its data modelling capabilities, and its ability to
identify patterns and trends previously undetectable by human
investigators highlight its transformative potential. Beyond that,
the use of AI for repetitive and resource-intensive tasks, allows
LEAs to work more efficiently with their limited resources and
lets police officers focus on and prioritise their most important
tasks.
Nonetheless, this comes at a cost, as certain applications of AI
in policing raise concerns over privacy, bias and discrimination.
There are concerns that these complex and somewhat opaque
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

systems may do more harm than good. Further deepening

the complexities of this new reality is the novel EU regulatory
framework – referred to as the EU AI Act1. This regulatory
framework introduces a set of new guidelines and standards
that will impact AI systems in use in the European Union.
While theAct aims to establish robust and ethical practices
across the board, law enforcement agencies will need to
review, and possibly modify, their existing and future AI tools
to ensure compliance. Certain applications of AI, such as
biometric identification – a longstanding law enforcement
practice - are expected to be strictly limited2. The EU AI Act will
also influence the development of future systems, making it
imperative for police to collaborate closely with AI researchers,
developers as well as ethical and privacy experts, to ensure
new systems are in line with regulatory guidelines.
Within this perspective, innovative methods such as
regulatory sandboxes and data spaces3 can serve as an

1 Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024
laying down harmonised rules on artificial intelligence and amending Regulations (EC)
No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and
(EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial
Intelligence Act)

8
adaptive mechanism in those cases in which the application
of AI-based technology could cause harm to the rights
and freedoms of data subjects. By creating controlled
environments where new AI tools can be tested and refined
using representative data, without real-world consequences,
law enforcement agencies can ensure that these tools
meet both operational and regulatory standards before
being deployed. This flexible approach allows for real-
time adjustments and fosters an innovative environment
of continuous improvement, positioning European law
enforcement at the forefront of AI-driven policing.

Objectives

The primary objective of this report is to provide an overview of

the benefits and challenges associated with the adoption of AI by
law enforcement. The report is intended to serve as an informative
resource primarily for LEAs operating across the EU, although the
foundational principles should be globally applicable. Nonetheless,
the report is a valuable asset for a diverse array of readers.
This includes policymakers, technology developers, academics,
civil rights advocates, and the general public, both within the
EU and globally. Through delving into the potential advantages
of AI integration, this report seeks to highlight how this rapidly
evolving technology can contribute to enhancing the efficiency,
effectiveness, and overall performance of law enforcement
operations, while upholding ethical and legal standards.

While the emphasis of this report lies in understanding the

applications, implications, benefits, and challenges of AI in law
enforcement, it does not explore the intricate technical details of
how AI algorithms or systems are developed, trained and operated.
This decision has been made to maintain the report’s accessibility
and to prioritise its primary objectives. Readers interested in a
deeper technical exploration of AI systems, their architectures, and
underlying mechanics are encouraged to consult specific technical
resources. Such sources can be found throughout the endnotes.

9
 Key takeaways for AI has the ability to significantly transform policing; from
advanced criminal analytics that reveal trends in vast amounts
law enforcement of data, to biometrics that allow the prompt and unique
identification of criminals.

The integration of large and complex datasets and Natural

Language Processing into policing applications allows for
the extraction of actionable insights from vast datasets,
improving resource forecasting and operational efficiency.
Simultaneously, these technologies can protect and uphold
individual privacy rights.

AI-driven tools, including in the context of OSINT and

SOCMINT, process unstructured data to provide real-time
insights and enhance the ability to tackle urgent situations
such as crimes against children and terrorism more effectively
and efficiently.

Technologies like machine translation are crucial to facilitate

international collaboration among law enforcement agencies.

The fusion of AI and biometrics can enhance criminal

identification accuracy while protecting the privacy of non-
relevant individuals.
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

Generative AI represents the next leap, moving from passive

analysis to active creation. For law enforcement, it offers a
treasure trove of possibilities. Yet, like any tool, its power lies in
its judicious and ethical application, balancing innovation with
responsibility.

The effective development and deployment of AI technologies

requires substantial technological infrastructure and expertise,
presenting significant challenges, particularly to smaller law
enforcement agencies.

Law enforcement agencies must navigate complex legal

and ethical landscapes while investing in training and raising
awareness amongst their staff to ensure appropriate data
handling and responsible data processing practices.

10
Compliance with the EU AI Act represents a crucial balancing
act, as it requires law enforcement to adhere to stringent
ethical, legal, and privacy standards, potentially necessitating
the reassessment of existing AI tools.

The EU AI Act challenges law enforcement agencies to

allocate additional resources and navigate the complexities
of compliance. This is especially relevant for those agencies
developing AI tools in-house, emphasising the need for a
responsible and ethical approach to AI integration in law
enforcement.

Police forces, which may already be utilising certain AI

systems, will face the challenging task of re-evaluating these
tools. Should any of these operational technologies fall within
the prohibited category set by the EU AI Act, they would
need to be deactivated, leading to potential challenges in
maintaining operational continuity.

Addressing bias in AI is paramount, with a need for systems

that are not only technically sound but also embody fairness,
justice, and impartiality, ensuring that data collection and
storage adhere to strict privacy guidelines.

Accountability, transparency and explainability, are essential

not only for ethical and responsible AI use but also to
ensure that evidence collected and analysed by AI systems
withstands scrutiny, respect the rights to a fair trial and is
deemed acceptable in court proceedings.

Regular audits of AI systems are essential to ensure

compliance with established privacy and data protection
standards, maintaining a balance between harnessing AI-
driven insights and safeguarding fundamental rights and
individual freedoms.

11
 Applications AI technology has the ability to completely transform policing; from
advanced criminal analytics that reveal trends in vast amounts of
of AI in law data, to biometrics that allow the prompt and unique identification
enforcement of criminals. This section explores some of the major applications
of AI within the field of law enforcement. Through this, we aim to
provide insight into the present and future capabilities that AI offers
policing, projecting a course for a more efficient, responsive and
effective law enforcement model.

Data analytics

The ability to analyse massive amounts of information and to then

promptly take efficient decisions, has become essential in the digital
age. In fields like law enforcement, decisions often need to be made
with limited resources in time-sensitive settings (e.g. during a police
raid, abductions or hostage-taking scenarios). Thus, the inability to
make sound decisions might have a profound societal impact and
negative consequences for citizens’ freedoms and rights.

At its core, data analytics entails the extraction of knowledge and

actionable insights from unprocessed and raw data. It provides
a way to identify patterns, trends, and links in vast datasets. The
advent of AI has significantly boosted the capabilities of traditional
criminal data analytics. The ability of AI systems to learn and adapt
based on data, including historical and other criminal data that are
available to law enforcement, allows criminal analysts to navigate
through, process and analyse vast amounts of information more
efficiently and accurately than any human ever could without this
type of technical assistance.
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

For example, using AI-driven analysis tools, investigators can

analyse millions of financial transactions and detect anomalies,
such as suspicious movements of funds to identify fraud4. For
law enforcement agencies, this translates to an enhanced ability
to analyse and understand crime patterns, detect links between
international investigations and develop tailored strategies to
specific challenges.

This transformative power of AI not only facilitates data processing,

but also enriches the quality of generated intelligence leads. For
instance, where traditional analytics might merely point out the
occurrence of a crime spike, AI-powered analytics could potentially
identify underlying causes, correlations between external and
unrelated events, or even subtle patterns that would go unnoticed
in manual analysis. It should be noted that this is typically done
in targeted use cases in the context of well-prepared and closed
datasets.

Furthermore, in crime areas involving digital devices such as

smartphones the amount of data to be analysed and acted upon
is massive and intricate. In these scenarios, AI-powered data
analytics becomes indispensable for effective analysis. Without the
assistance of AI, law enforcement agencies may face significant
challenges in deciphering vast arrays of data, leading to potential

12
oversights, prolonged investigations, and missed opportunities to
apprehend criminals. For example, simply going through the volume
of data generated by one single smartphone is impossible without
technical assistance.

The following sections will delve into the concepts of large and
complex datasets, OSINT/SOCMINT (Open Source Intelligence/
Social Media Intelligence) and Natural Language Processing (NLP)
and how they can reshape modern law enforcement practices.

LARGE AND COMPLEX DATA SETS

Police more and more often are facing the challenge of navigating
through large and complex datasets that cannot be easily managed
and processed using traditional data processing tools. Handling the
complexities such datasets require special techniques. Advanced
database management systems and scalable search solutions,
parallel processing5, and cloud computing infrastructures are often
employed to store, process and access massive data volumes.
Moreover, AI models including machine learning algorithms play a
crucial role in analysing and making sense of this data, especially
when human analysis would be too slow or inefficient.

The ultimate goal of navigating large and complex datasets is to

extract actionable insights. For law enforcement, this could mean
transcribing thousands of hours of audio files, extracting entities
such as names and phone numbers from text messages without
necessarily going through the content of the message, thus it
serves to restrict potential data protection violations and minimises
the amount of personal data processing. Other relevant applications
in policing include:

• detect patterns in criminal activity;

• identify correlations between different data types (like weather or

seasonal patterns and crime rates, e.g. rate of burglaries increase
during warmer months);

• forecast resource requirements based on past trends (e.g. a

police department is trying to determine how many officers it
should deploy in different precincts during different times of the
day and week).

This list is not exhaustive. New use cases of analysing large and
complex datasets within law enforcement will emerge as the
technology and criminal landscape evolves.

13
 Large and complex datasets in operations:
In 2020, the joint efforts of French and Dutch law enforcement, supported by Europol6, led to the
successful dismantling of the encrypted communications tool EncroChat. This operation not only
dealt a severe blow to criminal networks, but also demonstrated the crucial role of analysing large and
complex datasets in unravelling the intricate web of criminal activities at a global scale.
EncroChat, intended as a network for providing perfect anonymity, discretion, and no traceability to
users, served as a key tool for organised crime groups (OCGs) worldwide. Encrochat-enabled phones,
priced at approximately EUR 1000 each, offered features like automatic message deletion and remote
device wiping capabilities, making them indispensable for criminals seeking secure communications.
Since the dismantling, investigators managed to intercept, share and analyse over 115 million criminal
conversations, by an estimated number of over 60 000 users. User hotspots were prevalent in source
and destination countries for the trade in illicit drugs, as well as in money laundering centres7.
The success of this operation underscores the transformative impact of analysing large and complex
datasets. The vast dataset comprising millions of messages became a critical asset in dismantling
criminal networks. Through advanced analytics, law enforcement agencies were able to identify
patterns, connections, and hotspots, leading to the arrest of 6,558 suspects, including 197 High Value
Targets. The scale of this data-driven approach is evident in the seizure of criminal funds totalling EUR
739.7 million, the freezing of EUR 154.1 million in assets, and the confiscation of substantial quantities of
drugs, vehicles, weapons, and properties.
The EncroChat takedown serves as a paradigm for the effective integration of large and complex
datasets analytics in combating organised crime. Europol’s commitment and collaboration with various
stakeholders showcases the power of collaborative efforts and data-driven intelligence in disrupting
criminal activities around the world. It shows that this can be done while adhering to European data
protection and human rights standards, with consultation from the European Data Protection Supervisor
(EDPS). This operation stands as a testament to the evolving landscape of law enforcement, where
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

advanced analytics play a pivotal role in dismantling criminal networks and upholding the rule of law.

Making the most of what AI solutions have to offer does not rest
solely on the technology itself. Crucially, AI systems may only run
properly on appropriate, extensive technological infrastructure. This
requires significant budget and specific expertise to create and run,
which can be challenging to obtain, especially for smaller agencies8.

Furthermore, it is key to consider the implications of obtaining,

processing and analysing data and be mindful to make sound legal
and ethical choices at every step of the process9. Practically, the
EU has strict regulations and guidelines in place to ensure that
individuals’ privacy rights are protected, and that data are processed
fairly and lawfully for specified, explicit and legitimate purposes,
such as the General Data Protection Directive (GDPR) and the Law
Enforcement Directive (LED). Compliance with these regulations is
paramount, if at times restrictive for data analysis.

Lastly, the exchange of information among different agencies and

units within law enforcement can be a challenge10. Fragmented
data systems, information silos, and limited interoperability between
databases can obstruct the comprehensive analysis of large and

14
complex datasets. Collaboration and data sharing among agencies,
as well as the development and adoption of common standards
are vital to harnessing the full potential of data-driven insights, but
achieving this in practice often proves difficult.

To overcome these challenges, law enforcement agencies may

need to invest in training and infrastructure to enhance their data-
handling capabilities. They must also continually navigate the
complex landscape of legal and ethical considerations, ensuring
that their data practices remain both responsible and effective.
Finally, fostering improved communication and collaboration among
different agencies and units is critical for leveraging the full potential
of analysing large and complex datasets in law enforcement while
respecting the boundaries of data protection and ethics.

PREDICTIVE POLICING
Within law enforcement, decision-making processes are
increasingly reliant on intelligence derived from large and complex
datasets11. A recent advancement is “predictive policing”, employing
sophisticated statistical methods to extract valuable new insights
from vast datasets, for instance on crime records, events and
environmental factors identified in criminological insights. This
approach empowers police agencies to identify patterns related to
the occurrence of crime and unsafe situations, and to deploy forces
according to these insights to minimise risks.

Predictive policing12, leverages the capabilities of AI to enhance

the effectiveness and efficiency of policing activities. Primarily
implemented through rule-based machine learning models2,
predictive policing involves two fundamental steps: (a) data
collection and (b) data modelling (prediction). In the data collection
phase, police departments accumulate structured and unstructured
data from diverse sources, including historical crime data (time,
place, and type), socio-economic data, and opportunity variables13.
This information is supplemented in some cases with data from
probation and social services, among other relevant sources.
Subsequently, machine learning algorithms are employed to analyse
this data in training and prediction phases. The AI model identifies
patterns within historical data, associating indicators with the
likelihood of a crime occurring, and then generates risk scores as
predictive outputs.

Predictive policing manifests in two main types: (a) area-based

and (b) individual-based policing. Area-based algorithms identify

2 A system designed to achieve Artificial Intelligence (AI) via a model solely based on
predetermined rules. Two important elements of rule-based AI models are “a set of rules”
and “a set of facts” and by using these, developers can create a basic Artificial Intelligence
model. These systems can be viewed as a more advanced form of robotic process
automation (RPA). Rule-based AI models are deterministic by their very nature, meaning
they operate on the simple yet effective ‘cause and effect’ methodology. This model can
only perform the tasks and functions it has been programmed for and nothing else. Due to
this, rule-based AI models only require very basic data and information in order to operate
successfully (Source: https://wearebrain.com/blog/rule-based-ai-vs-machine-learning-
whats-the-difference/).

15
 connections between locations, occurrences, and historical crime
statistics to forecast the likelihood of crimes occurring at specific
times and places. For instance, they can predict increased crime
rates during certain weather conditions or at major sporting events.
Individual-based predictive policing anticipates persons most likely
to engage in criminal activities. This approach has gained traction
in various EU member states, including the Netherlands, Germany,
Austria, France, Estonia, and Romania, with others exploring its
potential implementation14.

Real-world applications:
The Dutch Police, developed and operationalised the Crime Anticipation System (CAS)15 to address
a range of crimes beyond initial targets, including domestic burglary, robberies, pickpocketing,
car burglaries, violent crimes, commercial burglaries, and bicycle theft. The system conducts
weekly analyses using both local and recent data, enhancing this with external information about
neighbourhoods and their inhabitants. This is further enriched by the police’s own insights into criminal
activities, local conditions, and data from statistics in the Netherlands. It aims to identify crime patterns,
such as a higher frequency of bicycle thefts in a specific area occurring between 9:00 p.m. and
midnight. With these insights, the police can allocate their resources more efficiently and tackle these
crimes more effectively, as reported by local news16.
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

___________________________________________________________________________________________

Image source: https://nl.wikipedia.org/wiki/Criminaliteits_Anticipatie_Systeem

Despite the potential benefits of predictive policing, concerns have

been raised globally by policymakers and human rights groups
regarding its potential to infringe upon fundamental human rights.
The EU AI Act attempts to address these concerns; the relevant
provisions will be discussed in Chapter 5.

16
In conclusion, predictive policing represents a transformative
approach to law enforcement through the integration of AI
technologies. As its implementation continues to evolve,
policymakers must navigate the delicate balance between
harnessing the potential benefits and addressing the ethical and
legal concerns associated with this innovative policing tool.

OSINT AND SOCMINT

A sub-category of large and complex datasets originates from
Open-Source Intelligence (OSINT) sources, especially in an era
where the internet’s data footprint is expanding rapidly.

After 2020, online traffic increased significantly. Enforced

lockdowns, a result of the global pandemic, contributed to surges
in internet users worldwide. This newly defined “normal” also paved
the way for a spike in cybercrime17 and led to a significant increase
in violent extremist propaganda and terrorist content online18. In the
vast expanse of the digital world, where cyber criminals act swiftly,
traditional OSINT methods often struggle, confronted with the
dilemma of ‘overwhelming data, limited time’19. Delving into large,
diverse, and unstructured data sets to extract valuable intelligence
demands significant resources, such as time, personnel, and money
– assets not always available to many law enforcement agencies.

The momentum is substantially shifting towards automation,

optimising the use of resource and enhancing precision in
decision-making. In the OSINT sphere, automation assists the
user in uncovering and leveraging previously unidentified sources.
Consequently, a rising number of global law enforcement agencies
are adopting automated OSINT tools for investigative purposes.
From investigating and reconstructing online criminal footprints
to probing web applications and detecting cyber threats on social
platforms, the applications of an automated OSINT paradigm are
limitless. Automated OSINT tools provide insights that strengthen
early-stage investigations, helping investigators shift from merely
reacting to actively preventing.

Moreover, as already discussed, a major challenge remains; dealing

with unstructured data. To meet this challenge, law enforcement
may use automated, AI-enabled, multisource OSINT and Social
Media Intelligence (SOCMINT) tools, adept at managing both
structured and unstructured data. Empowered by self-learning
machine learning models, these tools can reformat unstructured
data, support targeted open-source searches and investigations,
and offer real-time insights. Crucially, this must all be done at a
speed that exceeds the speed at which criminals can erase their
digital tracks.

Furthermore, Online Service Providers (OSPs) and Internet Referral

Units (IRUs) can harness the power of AI to detect and counteract
terrorist propaganda, disinformation, hate speech and illicit online
content20. Utilising advanced AI and machine learning algorithms,
they can analyse vast amounts of data at high speeds to identify
patterns, keywords, or visual content associated with extremist

17
 ideologies. Moreover, AI-enabled systems can be trained on
known propaganda materials to proactively spot new content that
shares similar characteristics and signal this to law enforcement
officers, ensuring a more responsive and efficient takedown of
harmful content before it spreads. It should be noted, under the
new Digital Services Act (DSA), OPSs are not only encouraged but
required to enhance their monitoring capabilities to ensure safer
digital environments21. The DSA mandates a higher degree of
accountability and transparency from platforms, pushing OSPs to
disclose their content moderation practices and outcomes22.

However, the use of AI for content moderation presents a complex

challenge, particularly regarding the balance between the right
to freedom of expression and freedom of thought, conscience,
and religion (articulated in Articles 10 and 11 of the Charter of
Fundamental Rights of the European Union) and the imperative to
counteract disinformation, hate speech, and illicit content online.
This balance is delicate, as the deployment of AI might inadvertently
curtail legitimate expressions under the guise of content
moderation, posing a threat to these fundamental freedoms.

NATURAL LANGUAGE PROCESSING (NLP)

Natural Language Processing, commonly abbreviated as NLP, is
a branch of computer science and linguistics that focuses on the
interaction between computers and human language23. It seeks to
enable machines to interpret and generate human language in a
meaningful and useful way. Research indicates that NLP methods
are employed by law enforcement and police departments across
various activities. These include administrative duties, forensic
investigations, analysis of crime data, converting speech to text
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

for reporting, and documenting criminal activities24. The vast

amount of text-based data—ranging from interview transcripts,
witness statements, online communications and social media
posts extracted within the framework of criminal investigations—
can be analysed swiftly and efficiently using NLP. This efficiency
is particularly crucial when quick insights are needed in real-time
situations such as abductions or hostage taking scenarios, during
or in the aftermath of terrorist attacks and in investigating child
abuse and exploitation cases. Key tasks performed by NLP in
policing include:

f Text classification25: An analyst who processes textual data

often marks crimes with keywords to help understand the
circumstances surrounding a particular offence, like if an
offender was under the influence of alcohol or drugs. As crime
is constantly evolving, these labels may not be complete. One
example of text classification is the assignment of different
labels for subgroups of money laundering.

f Clustering26: Unlike text analytics that rely on predefined

characteristics, clustering can help group similar crimes.
Clustering maps texts into a high-dimensional space so that

18
 similar texts are close to each other. In this task, no labels are
required. Additionally, clustering can also consider factors such
as time and location, to provide a holistic view of crime trends.
For example, in burglary scenarios, clustering could reveal
emerging methods, like hooking keys through letterboxes or
exploiting particular lock weaknesses.

f Text summarisation: Text summarisation is a method used

to produce a concise and accurate summary of lengthy texts,
preserving the overall meaning. In the realm of NLP, two
primary approaches27 are employed: (a) extraction-based
and (b) abstraction-based. Extraction-based summarisation
involves extracting a subset of words or sentences that
encapsulate the key points from the text, potentially resulting in
grammatical inaccuracies. On the other hand, abstraction-based
summarisation utilises advanced deep learning techniques
to paraphrase and condense the original document, similar
to human summarisation. By generating new phrases and
sentences that encapsulate essential information from the
source text, abstractive machine learning algorithms prove
valuable help in addressing the grammatical limitations
associated with extraction-based techniques. This technology is
instrumental in assisting law enforcement in their work analysing
extensive police reports and other information. This technology
can provide law enforcement with concise summaries that
capture crucial details without sacrificing accuracy.

f Machine translation: Automated translation systems facilitate

the conversion of text from one language to another. These
models take text in a designated source language as input and
produce the corresponding text in a specified target language as
output. Google Translate stands out as a well-known example
of such a mainstream application. Machine translation systems
play a vital role in law enforcement by enabling efficient analysis
of multilingual communication data. These systems expedite the
processing of large volumes of information, aiding investigators
in uncovering potential threats and identifying criminal activities
across language barriers. The technology enhances global
collaboration among law enforcement agencies, allowing for
smoother communication and information sharing during
international investigations. Additionally, automated translation
contributes to evidence collection by accurately translating
diverse forms of evidence, thereby reducing language bias.
Europol’s own Secure Information Exchange Network Application
(SIENA), the state-of-the-art communication tool that connects
LEAs from 51 countries and 14 international organisations, already
allows the possibility for end-users to translate text from their
native language to English in real-time. The machine translation tool
provides fast, accurate, and context-aware translations to break
down language barriers and enhance communication amongst this
rapidly evolving network of LEAs.

Real-world applications of NLP in policing are diverse and

continually evolving. In cybercrime units, NLP aids in analysing

19
 criminal communication, deciphering hidden meanings, or flagging
potentially harmful online content. For example, major concerns in
fighting cybercrime include spotting predatory communications,
identifying internet criminals, and preventing child abuse and
online grooming. NLP can be a game changer for policing in that
regard28. A subtask of NLP, Named Entity Recognition (NER), assists
analysts in labelling entities in crime reports, according to their type
such as persons, organisations and vehicles. This allows for more
refined crime grouping and analysis. In the context of burglaries, for
instance, NER could distinguish between different entry methods,
such as breaking a window versus tampering with a specific lock
type. Furthermore, when sifting through massive databases of
unstructured text, NLP tools can extract crucial information (entity
extraction), allowing the police to act upon critical situations such as
threats to life, promptly and efficiently.

In essence, NLP acts as a bridge between the highly context-

dependent human communications and the performance of
computational analysis, equipping law enforcement agencies with a
powerful tool in their digital arsenal.

Digital forensics

Digital forensics has emerged as a critical discipline in the realm

of law enforcement, given our increasingly digitalised world.
With vast amounts of information being stored, communicated,
and processed digitally, the ability to accurately investigate the
digital footprint of criminals is crucial for the police. Central
to the advances in digital forensics is the role of AI in modern
digital investigations. AI provides an advanced capability to sift
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

through vast data repositories, automating processes that would

traditionally take human experts extensive periods29. For instance,
while a human investigator might manually sort through thousands
of files, AI can rapidly categorise, filter, and highlight relevant
information based on predefined criteria or patterns (e.g. image
classification or hash3 values).

Several tools and techniques for data recovery and analysis have
been developed with AI components. These tools can recover
deleted files, access data from damaged devices, and restore
fragmented pieces of information into coherent formats. Their
efficiency lies in their ability to adapt and learn from each case,
improving accuracy over time.

A significant concern in the digital space is cybercrime detection.

Malicious activities, from hacking to phishing attempts, often
leave subtle traces or are masked in regular web traffic. AI
excels in identifying patterns and anomalies within this data30.

3 Hash values are akin to digital fingerprints for files. By running a file’s contents through a
cryptographic algorithm, a distinct numerical identifier – the hash value – is generated that
represents the file’s content. Altering the content in any manner would drastically change
this hash value. Presently, the MD5 and SHA-256 algorithms are the predominant methods
for generating these hash values.

20
By continuously learning from new data, AI models can identify
between regular network traffic and potential threats, even if the
malicious activities evolve or employ new tactics.

Decrypting data is another area where AI has shown promise.

Advanced encryption techniques can be serious obstacles for
investigators. While traditional decryption might involve brute-
force attempts4 or seeking encryption keys, AI can predict
potential encryption patterns or expedite the decryption process
by narrowing down possible encryption keys based on pattern
recognition.

Lastly, analysing digital footprints across devices and platforms

has become paramount, especially with the proliferation of
interconnected devices in the Internet of Things (IoT). A single
individual might interact with multiple devices daily, from
smartphones and laptops to smart home devices. AI can trace
these interactions, building a comprehensive digital profile that aids
investigators in understanding a subject’s connections, or even
propose additional elements for further analysis.

Digital forensics, fortified with AI capabilities, has transformed the

investigative landscape, offering unprecedented depth and speed
in analysing digital data. This not only amplifies the efficacy of
investigations, but also positions law enforcement agencies to
tackle evolving digital threats proactively.

Computer vision and biometrics

In this rapidly evolving landscape, computer vision and biometrics

have emerged as game-changers for law enforcement, both from
prevention and investigation standpoints. As cities and communities
are facing a surge of digital imagery from sources like CCTV
cameras to personal devices, it is essential to use this vast visual
data31 effectively. Coupled with biometric techniques that utilise
the unique physiological traits of individuals, these technologies
promise a new frontier in policing. The fusion of biometrics and
AI can deliver a blend of efficiency and accuracy, offering in-depth
insights to swiftly and effectively identify criminals while at the
same time protecting the privacy of non-relevant individuals. As
LEAs navigate the challenges and opportunities of the digital age,
computer vision and biometrics stand out as invaluable allies.

VIDEO MONITORING AND ANALYSIS

4 A brute force attack employs a method of trial and error to crack login credentials,
encryption keys, or locate concealed web pages. Attackers systematically try every possible
combination in the hopes of making a correct guess. Such attacks are carried out using
‘brute force’, which involves continuous and forceful attempts to break into private accounts.

21
 The advancement of imaging technology, coupled with AI including
ML developments, has transformed the realm of law enforcement.
Some of the potential applications for law enforcement include:

f Real-time processing and anomaly detection: Video monitoring

has evolved beyond passive observation. With the integration of
AI-driven algorithms, video feeds can be processed in real-time,
scanning for predefined patterns or anomalies. This capability
shines especially in security-sensitive zones. The system can
promptly notify security personnel of suspicious activities,
such as vehicles near sensitive locations, unattended objects
like a forgotten bag at a transit hub, or unauthorised entries.
Additionally, this real-time processing can be instrumental in
traffic management, instantly detecting accidents or disruptions,
facilitating immediate and informed responses.

f Public safety and event management: For events like public

celebrations, concerts, or festivals, the safety and well-being
of attendees are paramount. Safeguarding these events is
fundamentally different from day-to-day policing. Instead of just
traditional visual overviews, AI-enhanced video analysis can
provide detailed insights into the general flow of participants.
This allows for the early detection of potential areas of
congestion and aids in proactive planning. Furthermore, the
system can identify situations that might need attention,
ensuring that everyone can enjoy the event peacefully.

f Auto-reporting of incidents32: One of the standout features

brought by AI integration into video analytics is its ability to
autonomously report incidents. If predefined conditions or
scenarios are detected, such as public disturbances or potential
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

safety hazards, the AI system can automatically generate

detailed incident reports or/and send alerts to officers to assess
the situation. This not only speeds up the documentation
process but also ensures that even minor incidents, which might
be overlooked in manual monitoring, are accurately recorded and
addressed.

In essence, modern video monitoring and analysis amplify the

capabilities of law enforcement. Police officers are not just
observing - instead they actively understand and interpret the vast
amounts of visual data at their disposal. From enhancing real-
time traffic management to ensuring public safety at large-scale
events, AI-driven video analytics represent a transformative leap in
law enforcement capabilities, offering unprecedented speed and
accuracy in detecting and responding to incidents, thereby fostering
a safer environment for all.

IMAGE CLASSIFICATION
In the realm of computer vision, image classification is increasingly
emerging as a critical field. Essentially, AI tools trained to categorise
images based on the dominant content or objects they detect,
help LEAs overwhelmed with imagery to promptly and effectively

22
analyse this data. Image classification helps swiftly sort through
such data, categorising images into groups such as ‘suspicious’
or ‘non-suspicious’ or even organising them by different themes,
events, or timeframes. This streamlined approach significantly
expedites investigative processes.

The evolution of modern image classification tools, especially those

powered by machine learning algorithms, has allowed for the rapid
processing of vast volumes of data. Such tools not only segregate
images with minimal manual intervention, but also ensure no piece
of vital visual evidence goes unnoticed. Furthermore, the inherent
precision in these systems ensures accurate categorisation,
leading to more effective investigative outcomes. Beyond
traditional applications, image classification is relevant in various
law enforcement domains. For instance, during public events
or crowded locations, image classification can identify potential
threats or disruptions, helping law enforcement to take preventive
actions.

A notable scenario that highlights image classification’s potential

is its application in forensic investigations; analysing data extracted
from mobile communication devices. Frequently, these devices
store thousands of images, which makes navigating them daunting
and time-consuming. Additionally, this situation raises significant
concerns about data protection when processing personal pictures.
Through AI image classification, images extracted forensically
from mobile devices can be rapidly sorted, minimising the need
for manual review and the amount of personal data processed.
By focusing only on relevant images, investigators can not only
save valuable time, but also unveil crucial information that might
have otherwise been overlooked in the vast volumes of data while
fulfilling the principles of data minimisation, privacy-by-design and
security-by-design. In essence, image classification is shaping the
future of digital investigations in law enforcement, offering a blend
of speed, precision, and efficiency.

BIOMETRICS
In an age where personal identification and verification are of
paramount importance, biometric technologies have ascended
as key instruments in the toolkit of law enforcement. Biometric
technologies allow the identification of individuals, using their
unique physiological (e.g. facial features, fingerprints, iris
patterns) or behavioural attributes (e.g. gait5, handwriting).
Facial Recognition: The technique of using facial images for
criminal identification is as old as modern policing. Until the early
1960s, the procedure was primarily manual and relied on individual
perception and human capacity to recognise familiar faces.
However, advances in imaging technology and computer science33

5 Gait refers to the manner or pattern of movement of the limbs during locomotion over a solid
substrate. Essentially, it is the way an individual walks or moves. Gait analysis is often used
in medical, sports, and rehabilitation contexts to understand and address various issues
related to movement.

23
 allowed for Automated Facial Recognition (AFR)34; computer
algorithms now assist the police, and digital images captured
through various means have long replaced printed photographs.

This technology uses algorithms to extract and analyse certain

facial features from images or video to match and verify identities.
It has become an invaluable tool for law enforcement agencies.
For instance, the technology helps swiftly identify suspects by
comparing facial data collected within the course of a criminal
investigation against historical data or databases of criminals
available to the police. Additionally, it plays a crucial role in locating
missing persons and children by matching unidentified individuals’
images against databases of those reported missing. Moreover,
outside the context of law enforcement, facial recognition offers
enhanced security in controlled environments, eliminating the need
for traditional authentication methods like physical access control.

However, the rise of facial recognition has also been followed by

concerns. Notably, bias remains a topic of debate. Some studies
have indicated discrepancies in the system’s efficiency, particularly
when identifying individuals from specific ethnic backgrounds,
genders, or age groups35. Privacy and data protection stands out as
another significant concern. As facial recognition systems become
more ubiquitous, especially in public domains, they spark debates
about surveillance’s ethical bounds and potential misuse. Moreover,
the data repositories that fuel facial recognition – vast databases of
facial data – can be attractive targets for cyberattacks, emphasising
the importance of robust data protection measures.

In this context, it is imperative to distinguish between systems

used in real-time in public spaces (Live Face Recognition, LFR)
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

and systems used retrospectively (post-event facial recognition).

When used retrospectively, AFR helps the investigators to compare
images of unknown persons, such as someone caught on CCTV
footage, suspected of committing a crime or a mugshot of an
arrestee, against a reference database. This reference database is
typically supervised and lawfully stored, such as custody images or
images collected during criminal proceedings.

Rather than focusing on one individual in pre-recorded imagery,

LFR performs a real-time reading of all people passing a camera,
regardless of their capacity, and compares them against a pre-
determined closed watch-list of persons of interest. In some
scenarios, the system will discard images that triggered no results
immediately to avoid undue infringement of applicable data
protection laws. LFR applications pose significant challenges both
from a technical and a human perspective (system load, human
capacity and biases, e.tc.). Police forces in the UK and in some EU
countries have trialled LFR applications with varying degrees of
success.

f
FACE RECOGNITION IN POLICING: REAL-WORLD USE CASES

1. Identification of an unknown person

24
f Biometric identification technologies, particularly facial recognition,
play a crucial role in law enforcement for prompt and efficient
identification of unknown persons. Two primary scenarios within
this context:
f Solving cold cases: In the investigation of a murder case, CCTV
footage identifies a suspect, leading to a facial image search
against a database of known and unknown individuals. Initial
results are negative, but the image is stored. Two years later,
a biometric query triggers a match during another murder
investigation, ultimately linking the suspect to the earlier case. This
demonstrates the power of biometrics in solving cold cases over
time.
f Uncovering child exploitation networks: In another scenario,
police confiscate a child sex offender’s computer, initiating a
biometric analysis of extracted images. Matches with victims from
previous investigations help unveil a broader network of criminals
involved in child exploitation. This underscores the significant
role biometrics play in combating heinous crimes and protecting
vulnerable populations such as missing children.

2. Targeted searches of a known person

f Law enforcement relies heavily on targeted searches of known

persons to validate identities and assess potential criminal
involvement6. Various scenarios highlight the importance of
biometric technologies in this domain:
f Unmasking terrorist connections: A citizen provides anonymous
information linking a certain individual to serious crimes and
terrorism. Traditional searches with biographic data yield no
results, prompting a facial image search. This reveals a potential
match with a wanted terrorist, illustrating how biometrics can
enhance leads and aid in counterterrorism efforts.
f Uncovering criminal networks through mobile data analysis:
Forensic experts analyse a suspect’s smartphone, using face
recognition to cluster media and narrow down targets. Subsequent
searches against biometric databases of known or unknown
persons reveal potential contacts, leading to the unravelling of a
broader criminal network. This case demonstrates the synergy
between technology and human analysis.
f Combatting financial fraud networks:In cases of ATM fraud, law
enforcement employs facial recognition to connect a known
perpetrator with a collection of images of unknown fraudsters.
This targeted search helps assess the involvement of the known
perpetrator in additional criminal activities.

Fingerprints: Fingerprinting is one of the oldest and most trusted

biometric techniques in law enforcement. Each individual’s
fingerprint pattern, consisting of ridges, loops, and whorls, is unique

6 It is essential to clarify the circumstances under which biometric searches are allowed,
especially when a person’s identity is questionable or self-declared. Consideration should
be given to cases where inconsistencies in identification, such as self-declared or fake
identities, may necessitate biometric searches to ensure accurate identification and prevent
potential threats.

25
 and remains unchanged throughout life, making it a reliable means
of identification. Traditional fingerprint analysis relied heavily on
trained experts who would manually compare prints, which was
time-consuming and sometimes subjective36.

Fingerprints and AI3738:

The integration of AI into fingerprint could revolutionise this domain:

f Automated matching: AI-driven systems can sift through vast

databases of fingerprint records in mere seconds, providing
matches with a high degree of accuracy. This speeds up the
identification process considerably, and is especially useful in
scenarios where quick results are crucial.

f Enhanced detail recognition: AI algorithms can identify

and highlight minutiae (specific points on a fingerprint, like
bifurcations or ridge endings) with more precision than the
human eye, leading to more detailed and accurate comparisons.

f Latent print analysis: AI is particularly beneficial when dealing

with latent prints - fingerprints unintentionally left on surfaces.
These prints might be partial, smudged, or low quality. Advanced
algorithms can enhance such prints, fill in gaps, or even predict
missing portions based on recognised patterns, enabling better
matches from otherwise challenging samples.

f Learning and adaptation: One of the strengths of AI systems

is their ability to learn. As they process more data, these
systems refine their algorithms, becoming increasingly adept
at recognising patterns or anomalies. This continuous learning
ensures that the fingerprint analysis remains state-of-the-art and
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

adapts to new challenges or techniques.

f Integration/interoperability with other systems: AI-driven

fingerprint systems can be easily integrated with other digital
databases or biometric systems, such as those large-scale EU
systems available to internal security practitioners (SIS, VIS,
Eurodac e.tc.). This allows for multi-modal biometric checks and
comprehensive background verifications.

f Incorporating AI into fingerprint analysis not only boosts the

accuracy and speed of the process but also brings a level of
consistency and objectivity, minimising human errors or biases.
It amplifies the strengths of traditional fingerprinting while
mitigating its limitations, making it an indispensable tool in
modern forensic and law enforcement contexts.

Voice recognition: Every individual has a distinct voice pattern,

shaped by the anatomy of their vocal tracts and their unique way

26
of speaking. Voice recognition technology deciphers these minute
differences, converting spoken words into digital models that can be
compared against stored voiceprints. In law enforcement contexts,
this can be utilised to match voice samples from phone calls or
recordings, confirm identities in security systems.

Iris scans: The intricate patterns in the iris, the coloured part of
the eye, are as unique as fingerprints. Captured through a simple
photograph, these patterns offer a quick and non-intrusive means of
identification. Although iris identification technologies were initially
adopted for military applications such as biometric registration of
vulnerable populations in battlefields39, the adoption rates by law
enforcement gradually increase40.

Gait analysis: An emerging field, gait analysis studies the way an

individual walks. Even subtle differences in posture, stride, and
pace can be captured and analysed, offering a non-invasive way to
identify individuals, especially useful in scenarios where facial or
other recognitions are not feasible.

To distinguish between retrospective and real-time biometric

identification applications, one should acknowledge the role of
the latter in rapid response scenarios, especially its usefulness
in preventing terrorist attacks, locating missing children and
stopping or addressing serious crimes. However, one must also
recognise the challenges and ethical considerations tied to real-
time biometrics, which highlight the imperative need for responsible
deployment and a regulatory overview to ensure privacy and prevent
misuse.

Biometrics, with its myriad of modalities, stands as a testament to

the innovative fusion of biology and technology. In law enforcement,
it offers not just the promise of accurate identification but can also
pave the way for more efficient services that respect the dignity of
citizens while safeguarding their security.

BIOMETRIC CATEGORISATION
A further application of AI that holds potential for law enforcement;
that of systems that facilitate the categorisation of individuals
based on their biometric characteristics have become increasingly
important. These systems, whose application is fundamentally
different from systems used for identification, serve as invaluable
tools for both prevention and investigation.

The primary drive behind employing these categorisation systems

in law enforcement is to protect vulnerable segments of the
population. For instance, biometric categorisation aids in detecting
and redacting sensitive data from images, especially when it
concerns minors or victims of severe crimes like child abuse or
trafficking. Such technologies ensure that the privacy and dignity of
victims are preserved during investigative processes.

It is important to understand the boundaries of biometric

categorisation. Police do not leverage these systems to deduce
or infer special categories of personal data such as sexual or

27
 political orientations, religious beliefs, disabilities, or affiliations to
trade unions. Instead, the focus primarily rests on age and gender
estimation. Still, such estimations, especially when integrated into
high-risk systems, necessitate robust regulatory overview, ensuring
that the technology’s potential is harnessed responsibly and
ethically, especially considering data protection concerns.

For example, the challenge of handling vast amounts of video

content, some potentially containing disturbing imagery of child
abuse, underscores the urgent need for accurate age estimation
systems. Without AI tools adept at identifying minors in videos with
greater accuracy than human analysts, the task of manual review
becomes nearly unfeasible due to the extensive resources needed.
In this context, biometric categorisation tools, capable of quickly
and accurately categorising individuals based on objective features
such as their age, prove to be invaluable resources.

Alongside their utility, it is imperative to consider data protection

aspects. Ensuring the privacy and security of individuals, especially
minors, during the processing of such sensitive data becomes
paramount. Balancing the urgency of accurate content moderation
with robust data protection measures remains essential in
upholding ethical standards and protecting the rights and privacy of
individuals involved.

Improved resource allocation and strategic planning

The increasingly complex landscape of law enforcement

necessitates a strategic approach to resource allocation. As threats
evolve and cities expand, ensuring optimal use of resources—be it
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

personnel, equipment, or time—is imperative. Artificial intelligence

holds the potential to transform resource allocation from a reactive
approach to a proactive, strategic one.

Understanding the need for optimal resource utilisation: LEAs

operate often under constrained budgets and personnel limits. Yet,
they are expected to ensure the safety of expanding urban spaces
and tackle emerging threats. Ensuring every resource is utilised
optimally is not just about efficiency—it is critical for public safety
and trust.

AI-driven strategic planning: Beyond just daily deployments, AI

plays a role in longer-term strategic planning. For instance:

f Organising patrols: Instead of generic routes, AI can design

patrol routes that change based on the time of day, day of the
week, or known patterns of activity, ensuring officers are where
they are most likely to be needed.

f Emergency response: AI can help in planning rapid response to

emergencies by proposing optimal routes, analysing real-time
traffic data, or even forecasting potential secondary incidents or
threats.

f Public event security: Large public gatherings, from concerts to

sporting events or parades, can be security challenges. AI can

28
 analyse past events, crowd dynamics, entry/exit bottlenecks,
and even social media chatter to help design a comprehensive
security plan.

Evaluating the effectiveness of policies/strategies: AI does not

just offer tools for planning, it is also pivotal for evaluation. Post-
incident reviews can be analysed to determine the effectiveness of
deployments or anti-crime policies.

Were police officers positioned optimally? Did the AI outcomes

match the actual patterns? Such evaluations can feed back into
the system, ensuring continuous learning and reviewing policies/
strategies.

In conclusion, AI-driven resource allocation and strategic planning

elevate law enforcement’s ability to safeguard communities. By
turning vast amounts of data into actionable insights, and by
continuously learning from both successes and failures, AI ensures
that law enforcement agencies remain agile, proactive, and ever
adaptive to the evolving challenges of the modern world.

Generative AI

The frontier of AI does not lie in just the analysis of existing data
and information, but also in the creation of entirely new content.
Generative AI, a rapidly advancing domain, employs algorithms to
generate content, including texts, images and other forms of media.
These technologies learn patterns, structures, and intricacies from
vast datasets and then produce new data that adheres to the same
patterns. For instance, after analysing thousands of images of
cats, a generative model can create a new, synthetic image of a
cat that, while entirely fictional, looks indistinguishably real. Some
of the most prominent forms of Generative AI include Generative
Adversarial Networks (GANs) and Large Language Models (LLMs).

Generative Adversarial Networks (GANs) are a class of machine

learning frameworks where one neural network learns to generate
as realistic synthetic data as possible while the other neural network
learns to detect synthetic data. While the networks interact with
each other, both are continuously improving their performance
over time. GANs, more specifically the synthetic data generating
networks are widely used in image generation, video generation,
and increasingly in other domains like music.

GANs could offer law enforcement ways to evaluate the

performance of biometric systems without compromising
individual’s privacy41. For example, GANs can help generate
synthetic facial images, fingerprints, and other biometric data,
which can be in turn used instead of real data, where real
data may not be readily available, to assess the accuracy and
robustness of recognition systems across diverse populations
and conditions. Additionally, they enable the development of anti-
spoofing technologies to combat identity fraud4243. However, as
these technologies are adopted, it is crucial for law enforcement

29
 to balance the benefits with ethical considerations and privacy
protections, ensuring that the use of synthetic media supports
public safety while respecting individual rights.

Large Language Models (LLMs) refers to a form of generative AI

with applications in Natural Language Processing (NLP). These
models are designed to process and generate human language.
By being trained on extensive text datasets, they are capable of
executing various language-oriented tasks. These models mark a
substantial advancement in the way machines grasp and produce
human language with widespread implications across multiple
sectors such as technology, entertainment, education, among
others.

LLMs can offer substantial benefits to law enforcement agencies.

These include supporting investigators to probe into unfamiliar
crime areas, facilitating open source research and intelligence
analysis, as well as the development of technical investigative
tools44. Additionally, LLMs can help speed up numerous
administrative tasks, such as the writing of reports and the
summarisation of information. Nonetheless, the use of LLMs by
law enforcement would require a secure environment that can be
trusted with sensitive information, as well as thorough assessments
concerning safeguarding fundamental rights and mitigating
potential biases.

Despite their impressive abilities, LLMs have several limitations,

such as the propensity to produce factually inaccurate or illogical
content, commonly referred to as “hallucinations.” Retrieval-
Augmented Generation (RAG) emerges as a potential solution to
some of these challenges. While many LLMs rely primarily on pre-
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

existing knowledge or publicly accessible information to produce

text, RAG goes a step further by integrating information retrieval
mechanisms. This means that RAG models can actively search
for and incorporate relevant information from pre-determined,
authoritative knowledge sources, ensuring that the generated
content is not only coherent but also contextually accurate and
up-to-date. Organisations thereby have more influence over the
produced textual content, while users have a better understanding
of the process through which the LLM generates its response.

Exploring the capabilities of RAG in handling criminal investigations

datasets is essential, yet it must be approached in a methodical and
regulatory-compliant manner. In our current era, where information
is paramount, there is an unprecedented demand for innovative
methods to delve into and interpret complex data.

Generative AI represents the next leap, moving from passive

analysis to active creation. For law enforcement, it offers a treasure
trove of possibilities. Yet, like any application, its power lies in
its judicious and ethical application, balancing innovation with
responsibility.

30
 Despite the benefits for law enforcement, the integration of AI faces
Technological several technical constraints that challenge its effectiveness and
limitations and efficiency:
challenges Data quality and accessibility are fundamental to the effectiveness
of AI in law enforcement, but challenges arise from disparities in
data collection and storage practices across jurisdictions. These
variations result in inconsistent datasets that may be incomplete
or biased, compromising the integrity of AI outputs. Additionally,
existing data often lacks the granularity required for AI applications,
as it was not originally collected with AI in mind. For instance,
police reports, though informative, may not capture unreported
or undetected incidents, skewing AI training and outcomes.
Standardised data collection protocols, coupled with data cleansing
and enrichment processes are essential for creating comprehensive
and unbiased datasets. Moreover, integrating robust data protection
measures is crucial to safeguarding individuals’ privacy and
ensuring compliance with applicable data protection regulations.
By addressing these issues, AI reliability in law enforcement can
be improved, better reflecting and addressing the complexity of
criminal activity while upholding ethical and legal standards.

Integration challenges: Integrating AI with existing law enforcement

systems and data processing pipelines presents various technical
hurdles. The incompatibility between modern AI solutions and older
technological infrastructures can lead to significant integration
issues, affecting data exchange and operational efficiency. Bridging
this gap requires a dual approach: retrofitting legacy systems to
enhance their compatibility with AI technologies and designing
future AI solutions with a focus on interoperability and modular
integration.

Scalability and performance under different conditions: The

effectiveness of AI tools in law enforcement must be maintained
regardless of the scale of data or complexity of operational
scenarios. Variability in incidents and environmental conditions
tests the adaptability of AI systems. Addressing these challenges
necessitates the development of AI models that are not only
scalable but also versatile, capable of adjusting to different
data volumes and operational demands without compromising
performance.

Maintenance and technical support: The rapidly evolving nature of

AI technology demands continuous updates and maintenance to
safeguard efficiency and security. However, the requisite ongoing
technical support can strain the resources of law enforcement
agencies, particularly those with limited access to IT expertise.
Establishing dedicated support frameworks and leveraging
partnerships with technology providers could offer sustainable
solutions to these challenges, ensuring AI systems remain up-to-
date and effective.

Addressing these challenges is not straightforward and requires

an AI governance framework and a concerted effort from multiple
stakeholders. Collaboration between law enforcement agencies,

31
 technology developers, policymakers, and the community is
crucial to navigate these technological limitations. Through such
collaboration, innovative solutions can be developed, tested, and
refined to enhance the efficiency, reliability, and overall effectiveness
of AI applications in policing practices. Additionally, investing in
research and development, focusing on ethical AI use, and fostering
an environment of continuous learning and adaptation among law
enforcement personnel are key steps toward overcoming these
obstacles.

Ethical and social As the previous section demonstrated, Artificial Intelligence is

becoming an increasingly vital tool for policing across the European
issues in AI for law Union. Nonetheless, this brings forth a multitude of ethical and
enforcement social challenges that require meticulous analysis. This chapter
delves into critical areas of concern: the potential for data bias
and the subsequent implications for fairness; the fine line between
surveillance for security and the infringement of individual privacy;
the pressing need for accountability and transparency in AI
deployments, with an emphasis on the ‘black box’ issue. Moreover,
the chapter will discuss the potential for AI to either exaggerate or
mitigate human rights issues and discrimination within the realm of
law enforcement.

Data bias and fairness

Data is the core of any AI system, and the quality of the data directly
influences the outcomes produced by the system. Any skew in
data can unintentionally lead to unfair or biased outcomes. Fair and
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

unbiased policing is a foundational pillar of democratic societies,

and, therefore recognising and eliminating bias is of particular
concern to law enforcement.

Bias in data can emerge from numerous sources. Historical data45

used to train AI systems can embed longstanding societal biases,
reflecting past prejudices and discriminatory practices. For instance,
if a certain neighbourhood was historically over-policed due to
racial or socio-economic biases, an AI system trained on this data
might suggest that the area is more prone to criminal activity. Such
outcomes might create a feedback loop46, leading law enforcement
to continue over-policing that area, thereby finding disproportionate
numbers of crime and reinforcing the biases present in the data.

Beyond historical biases, there is also the challenge of

representational bias47. If data does not adequately represent
all segments of the population, the AI system can make flawed
predictions. Overrepresented groups can be disproportionately
affected. For instance, a study by the EU FRA48 found that offensive
speech detection algorithms, such as those for identifying hate
speech or harassment, had higher error rates for certain socio-
economic groups. A major contributing factor is the association of
certain terms with ethnic groups (e.g., ‘Muslim’, ‘gay’, ‘Jewish’), which
can cause the algorithms to mistakenly classify non-offensive

32
phrases as offensive. Since these terms are more frequently utilised
by the respective ethnic groups, there is an increased likelihood of
their content being wrongly flagged as offensive and subsequently
removed, due to their overrepresentation in the training data. On the
other side, groups that are underrepresented in the data may not
benefit from the same level of policing protection.

It is worth noting that there is not a universal agreement on the

precise definitions of fairness. Various interpretations exist. In some
instances, it is justified to use protected categories like gender
and age; for instance, an AI system that deducts information
about minors to ensure additional protection needs to be trained
with relevant sensitive data. As such, these situations should
be evaluated individually, and ultimately, humans must always
determine how to act on the information provided by the AI.

Privacy and surveillance

In law enforcement, striking the right balance between public

security and individual privacy has always been a challenge. As AI
integrates more deeply into policing methods, this balance becomes
even more delicate.

Historically, law enforcement agencies across the EU operate within

a robust legislative and regulatory framework. The introduction of
regulations such as the General Data Protection Regulation (GDPR)
and the Law Enforcement Directive (LED) underscores the EU’s
proactive stance on safeguarding data protection and individual
privacy rights. These regulations serve as foundational pillars
governing the intersection of technology and citizens’ rights, fortified
by robust enforcement mechanisms, human overview, and avenues
for redress. They are not merely legal frameworks but embody
comprehensive measures to ensure the responsible handling of
personal data, fostering transparency, accountability, and trust in
digital interactions.

While AI offers significant advantages for law enforcement, such as

the ability to process vast amounts of data and utilise biometrics
for rapid criminal identification and threat assessment, it also
brings with it complex challenges. Advanced technologies like facial
recognition systems can dramatically enhance efficiency. However,
without sufficient safeguards, such as human oversight to evaluate
their outputs, these technologies risk infringing on fundamental
rights, such as the right to private life and the right to personal
data protection (Art. 7 and 8 of the EU Charter of Fundamental
Rights). This could manifest as disproportionate surveillance of
innocent individuals or the potential for misuse targeting specific
groups, raising concerns about privacy and the necessity of such
monitoring.

As the world copes with the implications of AI and surveillance, the

EU, fortified by its stringent regulations, institutional ethos, and a
history of prioritising its citizens, is uniquely poised to shape a path
where technological advancements strengthen security without

33
 compromising individual rights. This coexistence can serve as
a global model, ensuring that technology remains a tool for the
improvement of society.

Accountability and transparency

Accountability and transparency serve as cornerstone principles

in democratic societies, ensuring that power structures remain
in service to the community and function with integrity. As AI
becomes a prominent tool within law enforcement, these principles
must be at the forefront to maintain public trust and ensure justice.

Despite the benefits the technology brings, one of the

primary concerns is the potential for decisions, predictions or
recommendations made by AI to remain unexplained or unjustified.
When the output of AI is used to support decision making in law
enforcement – be it biometric identification, or threat assessment–
it is crucial for both police officers and those affected by these
decisions to understand the rationale behind. Without this clarity,
the risk of mistrust, misuse, and potential injustices escalates.

In the EU, the demand for accountability and transparency is

not new. However, AI’s unique nature, where algorithms often
operate with layers of complexity beyond human comprehension,
introduces novel challenges.

There is a pressing need for mechanisms that make AI’s decision-

making processes interpretable, especially in high-stakes
environments like policing and criminal justice, not only in terms
of how relevant evidence are collected, processed and presented
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

before a court or tribunal, but also in a broader sense to ensure that

citizens can comprehend, engage with, and challenge the use of AI.

Ensuring accountability also entails setting clear responsibilities.

When an AI tool is used to generate recommendations or make
predictions, who is to be held accountable if there is an error
or if it results in injustice? Is it the software developers, the law
enforcement agency using the tool, or the overarching regulatory

34
 body? The definition of responsibility is vital to ensure that AI tools
in law enforcement remain both effective and just.

The Black Box Issue

In discussions surrounding AI transparency, a central and pressing concern is the puzzling ‘black box’
issue. At its core, the black box dilemma underscores the opacity inherent to complex AI algorithms,
with a particular focus on the complexities of deep learning models49. These Machine Learning (ML)
models are designed to emulate human information processing. Employing multiple layers of artificial
neurons connected to a network to extract advanced features from the input data, they bear the label
‘deep’. However, what raises profound questions is their capacity to make decisions or predictions
lacking of a clear, linear explanation for their rationale. Much like an opaque sealed black box, these
algorithms produce outcomes without exposing their inner workings to allow for an assessment of the
applied logic.
In the law enforcement domain, this opacity poses a significant challenge. When an AI-driven system
raises concerns about an individual’s potential threat or recommends additional patrol officers in
a specific area, it becomes imperative for law enforcement officers and the individuals affected by
such decisions to understand the underlying logic. The absence of this critical insight opens the door
to the potential for unchecked biases, errors, or misinterpretations, raising fundamental issues of
accountability and justice.

Resolving the black box predicament is not solely a technical challenge; it is a profound ethical
imperative. Innovative solutions, like explainable AI (XAI)50, are actively under development to bridge this
gap and render these algorithms more transparent and comprehensible. However, until such solutions
become universally accessible and standardised, the black box issue remains an indispensable focal
point in the ongoing pursuit for an AI-driven policing framework that is accountable, fair, and transparent
to all stakeholders involved.
It should also be mentioned, AI algorithms can be opaque due to their protected status as trade secrets.
Data controllers in these cases avoid sharing details of the inner algorithmic workings to protect trade
secrets and avoid system manipulation.

35
 Returning to the broader landscape, it becomes evident that for
AI to truly benefit law enforcement in the European Union and
maintain public trust, a rigorous commitment to accountability and
transparency is essential51. The development of frameworks to
explain AI’s decision-making processes, together with well-defined
regulatory standards and clarity in assigning responsibility, are
indispensable for establishing this balance.

Human Rights and Discrimination

In the EU, where human rights are deeply embedded in our

foundational values, integrating AI into law enforcement brings
forth several challenges. The primary concern is AI’s unintended
reinforcement or amplification of societal biases due to reliance on
historical data. As discussed, such biases can lead to unjustified
targeting of particular social groups, leading to disproportionate
policing.

Furthermore, AI’s predictive capabilities can mistakenly categorise

individuals based on broad data patterns. Such generalisations
might risk infringing on the fundamental principle of “innocent until
proven guilty,” raising valid concerns about the right to fair trial.

To foster a balanced integration of AI within this critical paradigm,

law enforcement has an array of options. Firstly, the significance of
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

undertaking comprehensive audits cannot be overstated. Every AI

system, before its active implementation in law enforcement, should
undergo an in-depth assessment. While the technical robustness
of these systems is essential, it is equally important to ensure their
conformity to the relevant frameworks such as the ethics guidelines
for trustworthy AI, introduced by the High-Level Expert Group on
AI52. By locating and addressing any inherent biases at this stage,
we can set the foundation for fair and unbiased AI implementations.

Equally crucial is the need to facilitate community engagement.

Certain communities frequently find themselves excluded from
the mainstream of technological advancements, often facing
unintended negative impacts as a result. Through fostering
continuous dialogue with these communities, law enforcement can
gather unique perspectives other than purely technical evaluations.
Proactive engagement not only improves trust but also ensures that
AI systems are deployed in a way that resonate with the broader
ideals of fairness, inclusivity, and justice.

Lastly, the dynamic nature of AI necessitates continuous monitoring

and evolution. Technologies evolve, societal norms shift, and new
challenges arise. In such a landscape, ensuring that AI applications
in law enforcement are subject to ongoing monitoring becomes

36
The EU Artificial essential. This iterative scrutiny and feedback enables real-time
adjustments, ensuring that AI-driven initiatives in law enforcement
Intelligence Act: consistently mirror and uphold the EU’s dedication to equal rights,
overview and context justice, and human dignity.

As the previous section highlighted, the increasing integration

of AI systems into various facets of policing, raises concerns
regarding their ethical, legal, and societal implications. The EU
Commission, recognising the transformative potential of AI in all
sectors of society, proposed in early 2021, a new legal instrument
to regulate the use of AI horizontally, while balancing innovation
with the protection of fundamental rights and societal values53. This
legislative proposal is referred to as the EU Artificial Intelligence
Act (EU AI Act). Following a lengthy consultation, the European
Parliament and the Council of the EU, reached an agreement and
adopted the Act which was published in the Official Journal on the
12th of July 2025 (Regulation (EU) 2024/168954). The Act will be
fully enforced gradually, within two years, with certain exceptions:
general provisions and the prohibitions will be enforced after 6
months, governance rules and obligations for general-purpose AI
models will apply after 12 months. Finally, the rules for AI systems
embedded in regulated products (Article 6(1)) will take effect after
three years.

As AI systems are more and more widely adopted, regulatory

frameworks are becoming increasingly important to clearly
outline what use cases are legally permissible. In the EU, this
legal framework has now been established via the EU AI Act. In
specifying how AI can be used in the EU, the AIA provides aims to
strike a balance between safeguarding core EU values, while still
allowing law enforcement to leverage the opportunities offered by
AI.

This section will delve deeper into the objectives, scope, and
key provisions of the EU AIAct, exploring its implications for law
enforcement agencies.

Objectives, scope and key provisions

The new regulations will be uniformly implemented across all

Member States, based on a forward-looking definition of AI,
ensuring a consistent application. According to the definition, ‘an
AI system is a machine-based system designed to operate with
varying levels of autonomy and that may exhibit adaptiveness after
deployment and that, for explicit or implicit objectives, infers, from
the input it receives, how to generate outputs such as predictions,
content, recommendations, or decisions that can influence physical
or virtual environments’ (Art. 3 of the EU AI Act). The definition
aligns closely with the work of international organisations working
on artificial intelligence, notably the OECD.

The scope of the Act is very wide, covering systems developed

with various approaches (machine learning, logic and knowledge-
based approaches, and statistical or Bayesian approaches) that can

37
 generate outputs such as content, predictions, recommendations,
or decisions influencing ‘environments they interact with’.

The main idea is to regulate AI based on its potential to cause harm

to society following a ‘risk-based’ approach: the higher the risk,
the stricter the rules. The ’unacceptable risk AI’ category prohibits
applications that are considered a clear and outright threat to
European values and fundamental rights, such as social scoring or
manipulation of human behaviour (e.g. toys using voice assistance
to encourage dangerous behaviour among minors). Meanwhile,
the “high risk AI” category includes specific systems which could
jeopardise people’s safety or infringe on fundamental rights; these
systems will not be prohibited but rather face stringent mandatory
requirements such as undergoing conformity assessments.

AI systems falling under ’limited risk’ will only be bound by basic

obligations, particularly in areas like transparency. For example,
in the case of using AI systems like chatbots, it is important for
users to recognise that they are engaging with a machine, enabling
them to make an informed choice on whether to proceed or step
back. All other AI applications, termed ’Minimal risk AI‘, can be
developed and used within the EU without any additional obligations
beyond existing legislation. On a voluntary basis, companies may
nevertheless commit to additional codes of conduct for these AI
systems.

The scope of the EU AI Act is extensive, encompassing systems

developed using various techniques, which includes machine
learning, logic and knowledge-based methods, as well as statistical
or Bayesian methods7. These systems are capable of producing
outputs like content, predictions, recommendations, or decisions
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

that affect the ‘environments they engage with55. Moreover, the Act
applies and imposes certain obligations to a wide range of actors,
including, providers (i.e. developers), deployers (i.e. users) and
distributors of AI systems (Art. 2(1) of the EU AI Act).

In the sections below, we will break down the main provisions

of the Act from a law enforcement perspective. While this is a
non-exhausting analysis, by understanding these, we can better
grasp how the EU plans to benefit from AI while making sure the
technology is used fairly and transparently for everyone.

PROHIBITED USES OF AI
In recognising the potential pitfalls and harms associated with
certain AI systems, the EU AI Act outlines certain AI practices that
are strictly prohibited (Art. 5). These prohibitions aim to prevent the
deployment of AI in ways that could cause potential harm, infringe
on individual rights, or undermine the foundational principles of the
EU. Applications that fall within this category include AI systems

7 Bayesian statistics is a method of data analysis that utilises Bayes’ theorem to revise existing
knowledge about model parameters using the information gained from observed data.

38
that manipulate human behaviour8, social scoring systems9 and AI
systems used to exploit the vulnerabilities of people (due to their
age, disability, social or economic situation).10

Moreover, the Act introduces a ban on real-time remote biometric

systems (RBI), when these systems are used in public-spaces.
Such systems are capable of capturing and analysing biometric
data (like facial features, iris patterns, fingerprints, voice patterns,
etc.) in real-time and from a distance, without the need for direct
interaction or physical contact with the individual being identified.
While these systems offer potential advantages for law enforcement
and security applications, their application in public spaces might
be considered intrusive, ‘evoke a feeling of constant surveillance
and indirectly dissuade the exercise of the freedom of assembly and
other fundamental rights’ (see Recital 18 of the EU AI Act). It should
be noted that the Act foresees some narrow law enforcement
exceptions (Art. 5(1)(h)). These exceptions will be discussed below.

Furthermore, Act prohibits the use of AI systems that allow the

biometric categorisation of natural persons based on certain
narrowly-defined attributes. These systems process biometric
data to deduce or infer their race, political opinions, trade union
membership, religious or philosophical beliefs or sexual orientation.
Such systems will thus be prohibited unless used to identify victims.
Filtering of datasets based on biometric data in the area of law
enforcement will still be possible (Art. 5(1)(g)).

One of the most critical additions has been the prohibition on

individual predictive policing systems. As discussed already, such
systems, designed to predict potential criminal activities based
on events, locations or persons, have raised concerns. There is
a fear that these systems might inadvertently reinforce biases,
leading to unwarranted surveillance or interventions. In response to
this, the EU AI Act introduces a partial ban of individual predictive
policing. The ban covers systems which assess or predict the risk
of a natural person to commit a criminal offence, based solely on
the profiling of a natural person or on assessing their personality
traits and characteristics. Use of AI systems that support the
human assessment of the involvement of a crime that has actually
occurred is allowed, as this is not considered a prediction but an
evaluation based on objective and verifiable facts directly linked to a
real criminal activity (Art. 5(1)(d))

While post-event remote biometric identification systems are not

outright prohibited, their high-risk categorisation mandates a third-
party conformity assessment process (Recital 125 of the Act).
Furthermore, the Act imposes additional obligations for users of
post-event RBI systems (Art. 26(10)). In particular, in the framework

8 AI applications that could potentially alter a person’s decision-making by exploiting

vulnerabilities or causing harm.
9 Systems that evaluate and classify the trustworthiness of individuals based on social
behaviour or known predicted personality characteristics.
10 Any AI application designed to use subliminal techniques that a person might not
consciously recognise but that could impact their behaviour.

39
 of an investigation for the targeted search of a person convicted or
suspected of having committed a criminal offence, the deployer of
an AI system for post-remote biometric identification shall request
an authorisation, prior to use, or without undue delay and no later
than 48 hours. The authorisation should be carried out, by a judicial
authority or an administrative authority whose decision is binding
and subject to judicial review. No such authorisation is needed if
the system is used for the initial identification of a potential suspect
based on objective and verifiable facts directly linked to the offence.
Moreover, the EU AI Act explicitly prohibits the untargeted use of
post remote biometric identification in law enforcement.

The EU AI Act takes a firm stance on preventing the creation or

expansion of facial recognition databases through the untargeted
scraping of facial images from the internet or CCTV footage (Art.
5(1)(e)). Untargeted scraping (e.g. extraction of data from a website),
is the collection of facial images without a specific, predefined
purpose. It goes beyond the necessary and proportionate use of
facial recognition technology, potentially amassing vast datasets
without clear objectives. This prohibition is designed to address
concerns related to mass surveillance and potential violations of
fundamental rights, particularly the right to privacy.

LAW ENFORCEMENT EXCEPTIONS TO PROHIBITED PRACTICES

Considering the specificities of law enforcement activities, the
co-legislators agreed on some exceptions to the prohibited AI
practices, as discussed above. Subject to appropriate safeguards,
these exceptions are meant to reflect the need to equip law
enforcement with all available tools to be efficient against modern
forms of crime, while also respecting the confidentiality of
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

sensitive operational data in relation to their activities. For example,

according to Article 46(2) of the EU AI Act, law enforcement or civil
protection authorities can put a specific high-risk AI system into
service urgently for reasons of public security or in the case of a
specific, substantial, and imminent threat to life or physical safety of
individuals. This can be done without prior authorisation, provided
that an authorisation request is submitted during or immediately
after the use of the system. If the authorisation is subsequently
rejected, the use of the system must be stopped immediately, and
all results and outputs from its use must be discarded.

Moreover, according to Art. 5(1)(h), the use of real-time remote

biometric identification systems in public spaces is possible
only for exhaustively defined law enforcement purposes. These
purposes include the targeted searches of victims, the prevention of
terrorist attacks and threats to life, and the localisation of criminals
suspected to be involved in serious and organised crime.

The circumstances under which law enforcement agencies are

allowed to use real-time RBI systems, are subject to specific
conditions (Art. 5(2)(a)):

40
f Specifically targeted individuals: The use is limited to
confirming the identity of specifically targeted individuals. This
implies that real-time RBI should not be used for indiscriminate
surveillance or broad identification purposes.

f Limited scope: The use of real-time RBI must be strictly

necessary and targeted. This includes limitations on the
individuals to be identified, the location, the temporal scope,
and being based on a closed dataset of legally acquired video
footage.

f Fundamental Rights Impact Assessment (FRIA): Law

enforcement authorities are required to complete a fundamental
rights impact assessment prior to using these systems. This
assessment would evaluate the potential impact on the rights
and freedoms of individuals.

f Authorisation requirements: The use of such systems in

publicly accessible spaces for law enforcement purposes must
be expressly and specifically authorised by a judicial authority
or by an independent administrative authority. While the EU AI
Act foresees exceptions to this rule11, this authorisation should
ideally be obtained prior to the use of the system (or within 24
hours).

f National laws: The exceptions for law enforcement use of real-

time RBI will be possible only if there is national law in place
explicitly foreseeing this, as outlined in the EU AI Act. As such,
Member States have the flexibility to decide on whether the
exceptions will be applicable in their country, introduce stricter
conditions, or even a horizontal ban of such systems.

f Notification of market surveillance authority: The relevant

market surveillance authority and the national data protection
authority should be notified of each use of the ‘real-time
biometric identification system’.

The RBI exceptions outlined in the EU AI Act are welcomed from a

law enforcement standpoint. These systems enable the targeted
and effective interventions, while avoiding disproportionate stop
and search measures based on race or ethnicity or any distinctive
physical characteristics. This strategic shift towards a more focused
use of technology not only enhances the ability of law enforcement
agencies to maintain public safety but also significantly reduces the
likelihood of discriminatory practices that have historically marred
policing efforts.

However, while these exceptions are seen as a positive

development, they also introduce a layer of complexity in the
broader context of AI tool adoption and application within law

11 Exceptions are allowed in urgent situations where obtaining prior authorisation is not
feasible, but even in these cases, the use must be restricted to the absolute minimum
necessary. If such authorisation is rejected, the use of real-time biometric identification
systems linked to that authorisation should be stopped with immediate effect and all the
data related to such use should be discarded and deleted.

41
 enforcement. While the Act is designed to ensure that relevant
technologies are used in a way that upholds fundamental rights
and fosters trust among the public, this may also slow down the
adoption process, as law enforcement agencies must navigate
through the additional regulatory requirements, ensuring that their
AI tools are compliant with the new standards.

This careful balancing act between leveraging AI for enhanced

law enforcement capabilities and adhering to the ethical, legal,
and regulatory standards set forth by the EU AI Act will likely
influence how AI technologies are upheld and implemented by
law enforcement agencies across the EU. The success of this
endeavour relies on finding a middle ground that allows for the
innovative use of AI for policing purposes while safeguarding
against the misuse of the technology in ways that could infringe
upon individual rights and freedoms.

HIGH-RISK AI SYSTEMS
The EU AI Act identifies certain AI applications in the realm of law
enforcement as ‘high-risk’ due to their significant potential to impact
individual rights, freedoms, and safety. By classifying these systems
as high-risk, the new regulatory framework mandates a set of
stringent requirements to ensure their ethical and responsible use.

Among the applications considered high-risk are biometric

systems used for unique individual identification tools like emotion
recognition and polygraphs aimed at assessing a person’s
trustworthiness or emotional state. Other high-risk applications
include some categorisation systems, as well as systems designed
to evaluate the risk of victimisation or offending by analysing the
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

likelihood of individuals becoming victims or perpetrators of crimes,

including human trafficking, domestic violence, or cybercrime.
Additionally, AI technologies employed to scrutinise the reliability of
evidence during criminal investigations or for profiling purposes in
detection and prosecution phases are subjected to these rigorous
regulations.

Notably, despite their initial categorisation as high-risk, the final text

of the EU AI Act excluded technologies for deepfake detection and
crime analytics, reflecting a nuanced approach to balancing the
benefits of these technologies against concerns related to privacy,
ethics, and misuse risks.

In light of the categorisation of certain systems as high-risk, users,

providers, developers and sellers of such AI systems, should follow
some strict rules. Every application, for instance, must undergo an
exhaustive risk assessment and mitigation process to understand
and counter potential hazards (conformity assessment). A thorough
fundamental rights impact assessment (FRIA) shall be performed
as well. The foundational data driving these AI systems must be of
the highest quality, not only to diminish risks but also to circumvent
any discriminatory outcomes and algorithmic bias.

42
THE FILTER MECHANISM FOR THE EVALUATION OF HIGH-RISK
SYSTEMS
The EU AI Act introduces a filter system to address concerns that
the classification of high-risk for certain AI applications might be
overly broad (Art. 6(3)). This system allows providers of AI systems
which could fall in the high-risk category, but which do not pose a
significant risk of harm to the health, safety or fundamental rights of
natural persons, to conduct self-assessments. The filter mechanism
focuses on AI systems tailored for less risky, specific functionalities,
including those for narrow procedural tasks, review or enhancement
of human-completed tasks, identification of decision-making
patterns, and preliminary task performance in critical assessment
preparation12.

For law enforcement agencies employing AI, this filter mechanism

could significantly benefit operations by reducing regulatory
burdens for AI systems deemed lower-risk. For example, AI tools
that assist in organising evidence or managing data, provided
they do not directly evaluate the reliability of evidence or profile
individuals. Another example would be AI systems that optimise the
logistics of evidence collection or systems used to organise and
cross-reference public records and case files, and aid in the initial
stages of an investigation, which could also fall within this filter
mechanism.

However, the operation of this filter system might pose practical

challenges, including potential difficulties in determining the
eligibility for exemptions and ensuring that the use of AI remains
within legal bounds. This could lead to uncertainties regarding the
scope of exemptions and the need for clear guidance and examples
from the European Commission to ensure that law enforcement
agencies can benefit fully, without compromising legal standards or
ethical considerations.

Implications for law enforcement agencies

The introduction of the EU AI Act poses various challenges and

implications for law enforcement agencies use of AI across the EU,
specifically regarding the deployment and utilisation of AI-driven
tools that include AI. Firstly, the Act’s explicit position on prohibiting
certain AI practices means there is an immediate imperative
to stop deploying these technologies. Police forces, which may
already be utilising certain AI systems, will now face the challenging
task of re-evaluating these tools. Should any of these operational
technologies fall within the prohibited category set by the Act, they
would need to be deactivated, leading to potential challenges in
maintaining operational continuity. This poses the question: how

12 The European Commission is tasked with developing guidelines for applying these filters,
aiming to clarify and simplify the compliance process for AI system providers. It should be
noted that AI systems used for the profiling of natural persons should always be considered
as high-risk applications.

43
 will the transition be managed for legacy systems under the new
regulations?

Moreover, the process of conformity assessments for systems

deemed high-risk by the EU AI Act will undoubtedly be intricate and
time-consuming. LEAs will be required to comprehensively assess
these systems against the stipulations set by the new regulation.
In many instances, this could entail considerable modifications
to existing systems to ensure alignment with the new standards.
Consequently, not only does this suggest potential changes to
software, but it also highlights the need to allocate additional
resources, in terms of both finance and staff.

Furthermore, the influence of the Act is not restricted to only

the newly deployed AI systems. Considering the dynamic nature
of AI, its continuous evolution and updates, systems that are
already in operation will also be subject to these regulations. The
evolving nature of AI means that LEAs will be in a perpetual cycle
of reviewing and modifying, ensuring that their systems, even if
previously compliant, remain in line with the regulations, especially if
updates alter their functions or associated risks.

A particularly challenging scenario emerges for LEAs that have

taken the initiative to develop AI tools internally. These agencies
will confront the dual responsibility of ensuring compliance both
as users and as developers. This essentially indicates a substantial
investment in guaranteeing that every stage of the process, from
development, data collection, training, to deployment, is in strict
adherence to the EU AI Act requirements.
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

44
Additional safeguards in the context of RBI for law enforcement.
Given the sensitive nature of RBI deployments and the implications on privacy and other fundamental
rights, the EU AI Act foresees additional safeguards on the use of such systems. According to the
regulation, law enforcement agencies using such systems, shall ensure that no decision that produces
an adverse legal effect on a person may be taken solely based on the output of these post remote
biometric identification systems (See Art. 26 (10) of the EU AI Act). Therefore, the new regulation
mandates an additional layer of verification and confirmation.
Achieving this requires a combination of technological, procedural, and legal safeguards such as human
validation, implementing multi-modal biometric systems, establishing appropriate confidence thresholds
and educate human operators on the capabilities and limitations of biometric identification technology.
Moreover, the EU AI Act considers an enhanced human overview as a requirement for those systems
so that no action or decision may be taken unless the output of the RBI system has been separately
verified and confirmed by at least two natural persons, with the necessary competence, training and
authority (Art. 14 (5)).
The requirement for a separate verification by at least two natural persons shall not apply to high risk AI
systems used for the purpose of law enforcement, migration, border control or asylum, in cases where
Union or national law considers the application of this requirement to be disproportionate. The mandate
of the so-called 4-eyes principle, reflects the peer-review process which features prominently in the
forensic sciences.
Nonetheless, adhering to this principle for rather basic investigative measures such as a criminal
identification, presents potential challenges for law enforcement agencies such as operational efficiency,
availability of resources, subjectivity in verification, expertise and training timeliness in critical scenarios
and others.

As the model EU law enforcement Agency, Europol is leading efforts

aimed at demonstrating compliance with the provisions of the EU AI
Act. In that direction, Europol, jointly with selected partners13 has co-
developed CC4AI, a compliance checker tool for the AI Act. CC4AI is
a step-by-step guide designed to support LEAs to evaluate whether
existing or future AI applications used in policing meet the criteria
set by the new regulatory framework.

With the CC4AI initiative, Europol seeks to contribute to ensure

a harmonious adoption and implementation of the EU AI
Act requirements across LEAs in the EU, thereby preventing
inconsistencies that could potentially impede police collaboration in
the future. Access to CC4AI will be offered freely to internal security
agencies.

While the new regulation offers a comprehensive framework geared

towards maximising the advantages of AI without compromising
ethical standards and the safety of the public, it also introduces a
complex set of challenges for LEAs. Successfully navigating this
regulatory terrain will necessitate a profound understanding of AI’s

13 EU JHA agencies (CEPOL, Eurojust, EUAA and EU FRA) from the EU Innovation Hub for
Internal Security and the Centre of Excellence in Terrorism, Resilience, Intelligence and
Organised Crime Research (CENTRIC)

45
 technological aspects, coupled with an increased awareness of its
legal dimensions within the realm of law enforcement.

Innovation and regulatory sandboxes

Central to the EU AI Act and its innovative approach is the

establishment of coordinated AI ‘regulatory sandboxes’ (Art. 57-60).
These controlled environments are designed to allow developers to
test and refine innovative AI products and services within a secure
environment. The Act’s emphasis on regulatory sandboxes is of
great importance to law enforcement agencies. As extensively
discussed, AI innovations, ranging from biometric identification
to advanced data analysis, offer substantial benefits to LEAs.
Regulatory sandboxes offer a unique opportunity for these agencies
to explore these technologies, understand their implications, and
identify any potential issues before implementation in real-life law
enforcement contexts.

Furthermore, the Act encourages a harmonised approach across

Member States, aiming to eliminate regulatory disparities. This
ensures that LEAs across the EU can uniformly benefit from
AI advancements, promoting consistent and effective use of
technology in law enforcement operations.

Another critical aspect addressed in the EU AI Act is the

intersection between sandbox activities and the EU’s stringent
data protection regulations, including the General Data Protection
Regulation (GDPR) and the Law Enforcement Directive (LED).
The Act acknowledges the importance of balancing the need for
experimentation with adherence to data protection principles,
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

emphasising the need for transparency in the deployment and use

of AI within sandboxes, especially in sensitive domains like law
enforcement where public trust is crucial.

The provisions within the EU AI Act for regulatory sandboxes

represent a forward-thinking approach, allowing law enforcement
agencies to harness the potential of AI technologies. By addressing
key concerns around liability, regulatory harmonisation, and
data protection, the Act ensures that innovation can happen in a
manner that is responsible, transparent, and aligned with the core
values and rights upheld by the European Union. Already in 2022,
the EU legislators had anticipated the use of sandboxes to allow
Europol to process personal data for the purpose of its research
and innovation projects to train, test and validate algorithms, in a
separate, isolated and protected data processing environment (Art.
33a of Europol Regulation)14.

14 REGULATION (EU) 2022/991 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL

of 8 June 2022 amending Regulation (EU) 2016/794, as regards Europol’s cooperation
with private parties, the processing of personal data by Europol in support of criminal
investigations, and Europol’s role in research and innovation (Source: https://eur-lex.europa.
eu/legal-content/EN/TXT/HTML/?uri=CELEX:32022R0991#d1e2572-1-1)

46
 As highlighted throughout the report, the fusion of AI and policing
Balancing the promises enhanced efficiency, new or improved capabilities,
benefits and and resource optimisation. However, it also brings critical issues
to the fore, such as potential biases, potential infringements of
restrictions fundamental rights and freedoms, and questions of accountability.

This section looks at the primary concerns and corresponding

strategies, emphasising the significance of ensuring fairness,
safeguarding individual rights, upholding stringent standards of
accountability, and fostering an environment that promotes both
innovation and stringent regulatory adherence.

Addressing concerns of bias and discrimination

Even though the foundational issues surrounding data bias and

fairness in AI for law enforcement were elaborated in section 3.1,
addressing these concerns goes beyond just the technicalities of
data. The implications of bias and unfairness in AI systems resonate
deeply within societal structures, trust dynamics, and the broader
ideal of democratic governance.

In law enforcement, where the consequences of decisions can be

life altering and the goal is to contribute to fairness and justice,
AI tools need to be more than just technically sound; they should
embody the principles of justice, fairness, and impartiality.

The challenge thus extends from ensuring fairness at data level,

to ensuring the equitable application and interpretation of AI
outcomes in real-world scenarios. By adhering to these principles,
law enforcement agencies can ensure that AI-processed evidence
meets the rigorous standards required for court acceptance,
thereby maintaining the integrity of the judicial process.

To address this multifaceted challenge, there is a need for a multi-

disciplinary approach. While an AI system might be trained on
unbiased data and might be using a fair algorithm, the way its
outcomes are interpreted and applied by law enforcement officials
could introduce bias. Ensuring that officers understand AI outputs
and follow clear guidelines on how to act on them can prevent
misapplications. For example, the output of a facial recognition
system is merely an indication that two persons present similarities
to a certain degree. Without human decision and corroboration of
these findings with additional information (such as fingerprints or
biographical data), this output should not be used as a probable
cause for arrest.

Moreover, human oversight and transparency mechanisms needs to

be established. Engaging communities in the evaluation of AI-driven
policing tools can be an effective way to gather trust and ensure
that the systems are in line with societal values. Recursive cycles
with communities can also help refine these tools to ensure they
are more aligned with community needs. Additionally, continuous
training of law enforcement personnel on the ethical dimensions of
AI is imperative. Officers should be equipped not just to use these

47
 tools but also understand their implications on society, ensuring
that the technology is used responsibly.

In conclusion, while the technical aspects of bias and fairness are

undeniably critical, the broader challenge lies in integrating AI into
the ethos of law enforcement in a manner that upholds the values
of justice, equity, and community trust.

Safeguarding privacy and data protection

As AI systems gain traction in law enforcement, concerns about

privacy and data protection rise concurrently. The use of AI-driven
analytics, predictive policing, and biometric identification systems
underscores the acute necessity to protect individuals’ personal
data and privacy rights.

Key to addressing these challenges is the establishment of a robust

data protection and AI governance framework. Data collection
should be governed by the principles of lawfulness, fairness, and
transparency, ensuring that only necessary and relevant data
is processed. Furthermore, data storage duration, especially
concerning personal information, must be limited strictly to what
is necessary for the purposes for which the data is processed.
Access to stored data should be restricted to authorised individuals
or entities according to applicable laws to prevent unauthorised
access.

Furthermore, encryption and other appropriate technical and

organisational measures must be implemented as standard
procedures to ensure the security and confidentiality of stored data,
thereby protecting against breaches and unauthorised access. For
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

AI systems that necessitate continuous data inputs, it is imperative

to adhere to the principle of data minimisation, collecting only the
data that is strictly necessary. This aligns with the requirements
outlined in the General Data Protection Regulation (GDPR) and the
Law Enforcement Directive (LED), which emphasise the importance
of limiting the processing of personal data to what is essential for
the intended purposes.

Equally important is to guarantee the individual’s right to

information. Citizens should be informed about the nature of data
being collected, its purpose, and the extent of its use. Moreover,
individuals should have the right to access, rectify, or even erase
their data under certain circumstances, in line with applicable data
protection laws.

Ensuring appropriate human overview is essential. While AI systems

are capable of processing and analysing vast amounts of data,
human intervention is necessary to ensure that the interpretation
of data aligns with legal and ethical standards. Conducting
regular audits of AI systems can provide additional assurance
of compliance with established privacy and data protection
regulations.

48
 Engaging with the public and fostering a culture of transparency
can also help build trust. Open dialogue about the scope and
limitations of AI in law enforcement, its implications on privacy, and
measures taken to safeguard data will foster community trust and
collaboration.

To summarise, as AI reshapes modern policing, safeguarding

privacy and data protection is of utmost importance. It is a balance
of harnessing the potential of AI-driven insights while ensuring
that the fundamental principles of privacy and individual freedoms
remain intact.

Future outlook and Potential for technological advancements

recommendations In the swiftly evolving landscape of AI, the horizon for technological
advancements remains vast. The intersection of AI with law
enforcement is particularly ripe for innovation, promising to reshape
traditional paradigms of policing and security.

f Quantum Computing: With quantum computing nearing

development, the potential for handling complex computations
of specific mathematical problems at unparalleled speeds
could introduce new digital forensics and decryption methods.
Quantum machine learning can improve computational speeds
and performance of AI models at unprecedented levels. A recent
report56 published by Europol’s Observatory sheds light into how
quantum computing and quantum technologies will impact law
enforcement and what should be done to prepare.

f 6G connectivity: 6G is expected to become one of the first

AI-native networks, integrating AI directly into the networking
infrastructure. This integration will empower the network to self-
learn and self-manage, enhancing its autonomy and reducing
operational costs57. This will likely allow for even faster data
transfer rates, facilitating real-time secure communication tools
for law enforcement, ensuring that officers in the field have
instant access to crucial information, enhancing on-ground
efficiency and safety. Consequently, the increase in data volumes
will likely require a new array of AI tools and methods to analyse
this data.

f Automated Drones and Robotics: The future might hold a more

widespread use of AI-powered drones and robots for search and
rescue operations, or even securing large events, providing aerial
insights or on-ground assistance without risking human lives.

f AI chips: The further development and integration of AI chips

will significantly accelerate the progress and capabilities of AI
technologies in the future. Specialised AI chips are designed
to efficiently process AI and machine learning tasks, offering
faster computation speeds and reduced energy consumption
compared to general-purpose processors. Further advancement
in this area will enable more complex and sophisticated AI

49
 models to be trained and deployed, enhancing the performance
of AI applications in law enforcement.

f Edge computing: By bringing computation and data storage

closer to the location where it is needed, edge computing has
the potential to revolutionise the future of AI. Edge computing
can enable faster, more efficient, and real-time AI processing
capabilities, which would reduce the dependency on cloud-
based services and data centres. This can help minimise latency,
bandwidth use, and the potential for data privacy breaches. For
AI applications, this means the ability to process and analyse
data on-device in real-time, which is crucial for applications
which require instant decision-making such as autonomous
vehicles, IoT devices, and smart city technologies. This could
have a significant impact on law enforcement, by enabling a
more seamless integration of AI at various operational levels.
Edge computing could facilitate use cases ranging from facial
recognition and mobile command centres to smart wearables,
improved sensors, and enhanced deployment of unmanned
systems, such as drones.
As these technological advancements come to the fore, it is
imperative for law enforcement agencies to stay abreast, adapt, and
integrate these tools responsibly. While the potential is significant,
ensuring that these technologies are employed ethically and in line
with the principles of justice and fairness will be paramount. The
balance between harnessing technology and safeguarding rights
will dictate the trajectory of AI’s role in the future of policing.

Building public trust and acceptance

AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

Public trust and acceptance are cornerstones for the successful

integration of AI technologies into law enforcement. Without the
collective confidence of the public, even the most ground-breaking
advancements risk facing resistance, potentially impeding their
effective utilisation. Investing in community engagement, education,
and feedback mechanisms can significantly enhance public trust
and cooperation in AI technology, ultimately leading to a more
informed and supportive community. For example:

f Community engagement: Regularly engaging with the

community can provide valuable insights into their concerns and
expectations and foster mutual understanding. Hosting public
forums, workshops, or open debates can help in demystifying
AI, addressing misconceptions, collaboratively identifying areas
of improvement. This dialogue is also essential to effectively
navigate the trade-offs between privacy and security, especially
in the context of high-risk AI applications in law enforcement.

f Education and awareness: Investing in public education

campaigns about the benefits and limitations of AI can reduce
fear and scepticism within the society. When people are
informed about the positive impacts, such as reduced crime

50
 rates or quicker response times, they are more likely to embrace
the technology.

f Feedback mechanisms: Establishing channels where the public

can voice their concerns, provide feedback, or even report
potential misuse of AI could instil a sense of participation and
co-ownership in the technology’s evolution.

Building public trust is an ongoing process, demanding consistent

efforts and open dialogue. As AI continues to shape the future of
law enforcement, prioritising public trust and acceptance will not
only validate technological progress but also ensure that society
moves forward cohesively, with technology acting as an enabler
rather than a divider.

Strengthening collaboration and knowledge sharing

within LEAs

In this dynamic ecosystem, collaboration and knowledge sharing

is of utmost importance. By fostering an environment of collective
intelligence and open dialogue, law enforcement agencies can
ensure that AI’s potential is recognised in a manner that resonates
with the collective good. This includes:

f Inter-agency collaborations: agencies across regions and

countries can come together to pool resources, share insights,
and co-develop AI tools tailored for diverse scenarios. Such
collaborative endeavours can streamline efforts, reduce
redundancies, and lead to more universally adaptable solutions.
Such an initiative is the aforementioned EU Innovation Hub for
Internal Security - a collaborative network of innovation labs
that works to provide the latest innovation updates and effective

51
 solutions to support the work of internal security actors in the
EU and its Member States, including justice, border security,
immigration and asylum and law enforcement practitioners.

f Partnerships with academia and the industry: By building

alliances with universities and technology companies, law
enforcement agencies can tap into cutting-edge research,
methodologies, and tools. This collaborative spirit can accelerate
the development and refinement of AI systems, ensuring they
remain at the forefront of innovation.

f Open-source AI: Promoting open-source AI projects can

facilitate access to advanced tools, enabling even smaller law
enforcement entities to leverage the power of AI. Such initiatives
can also foster a community-driven approach to development,
enhancing the quality and applicability of tools.

f AI Literacy: Enhancing AI Literacy among law enforcement

staff, in alignment with the EU AI Act, is pivotal for fostering
an informed and ethical approach to AI integration in policing.
Law enforcement personnel should become aware of the
operational, ethical, and legal dimensions of AI technologies,
ensuring they can effectively deploy AI tools while addressing
concerns related to bias, privacy, and accountability. Similarly, it
is crucial to engage the public in transparent discussions about
AI’s role in law enforcement, clarifying its benefits, limitations,
and regulatory oversight as mandated by the new regulation.
By promoting a deep understanding of AI’s capabilities and the
legal frameworks governing its use, this effort seeks to build
trust, enhance collaboration between law enforcement and
communities, and ensure AI’s deployment aligns with societal
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

values and fundamental rights.

f Centralised knowledge repositories: Establishing centralised

databases or platforms where agencies can share case
studies, research papers, toolkits, and best practices can serve
as invaluable resources. Such repositories can ensure that
knowledge is not just created but also made accessible to those
who need it. A notable example is Europol’s Platform for Experts
(EPE)15 and Europol’s Tool Repository (ETR)16. The latter was
created by the Europol Innovation Lab as a participatory platform
for law enforcement agencies around Europe, to share innovative
tools based on cutting-edge technology. The former, is a secure
environment for specialists in a variety of law enforcement areas,
including AI, enabling them to share – within their respective
communities - knowledge, best practices and non-personal data
on crimes. For the time being, there are over 18,000 members

15 Read more about EPE here: https://www.europol.europa.eu/operations-services-and-

innovation/services-support/information-exchange/europol-platform-for-experts
16 ETR is a secure and LEA-exclusive online platform to share non-commercial, cost-free
software developed by LEAs and research and technology organisations. ETR was
designed to avoid duplication of effort among LEAs. It allows all police forces in Europe to
benefit from innovative tools and become more efficient in their work, thus enhancing the
protection of EU citizens. AI tools available through ETR have already supported several
operations across Europe, resulting in the arrests of organised criminals and the rescue of
victims of human trafficking.

52
 from more than 100 countries interacting and collaborating with
each other in virtual communities.

f Engagement with civil society: Collaborating with civil actors’

society and community organisations can provide unique
perspectives on the ethical and societal implications of AI in
policing.

Conclusion The integration of AI into law enforcement, particularly within

the European Union, presents a paradigm shift with profound
implications for both operational efficiency and ethical
considerations. This report assessed the multifaceted applications
of AI in policing, ranging from data analytics and predictive policing
to digital forensics, computer vision, and biometrics. The report
highlighted the transformative potential these technologies hold for
enhancing public safety and operational effectiveness.

However, this technological evolution does not come without

challenges. The ethical and societal dimensions, including concerns
over data bias, privacy, accountability, and human rights, are critical.
The European Union’s proactive stance through the development
of the Artificial Intelligence Act underscores a commitment to
balancing innovation with ethical considerations, aiming to foster
an environment where AI’s benefits in law enforcement can be
harnessed responsibly.

The report emphasised the importance of addressing bias and

discrimination, safeguarding privacy and data protection, and
ensuring accountability and legal compliance. It outlined strategies
such as fostering regulatory sandboxes, which allows for the safe
exploration and development of AI technologies within a structured
framework that respects EU values and fundamental rights.

Looking forward, the trajectory of AI in law enforcement is poised

for significant technological advancements, including quantum
computing and 6G connectivity, promising to further enhance the
capabilities of law enforcement agencies. However, the successful
integration of these technologies hinges on building public trust
and acceptance, emphasising transparency, and strengthening
collaboration and knowledge sharing within the law enforcement
community.

In conclusion, the report called for a balanced approach to AI in law

enforcement, where the benefits of technological advancements
are leveraged to enhance public safety and operational efficiency,
while simultaneously addressing the ethical, legal, and societal
challenges. The EU’s regulatory framework, including the Artificial
Intelligence Act, provides a solid foundation for this endeavour,
ensuring that AI’s integration into law enforcement aligns with the
EU’s core values and fundamental rights. The future of AI-driven
policing, therefore, lies in the harmonious integration of innovation
with regulation, guided by principles of fairness, accountability, and
transparency.

53
 AI Glossary ACCOUNTABILITY: The responsibility and explainability for the actions
and decisions made by Artificial Intelligence systems. In the context
of AI-driven policing, it involves ensuring that the use of AI in law
enforcement adheres to ethical standards and legal regulations.

AI GOVERNANCE: The framework and set of rules and regulations that

guide the development, deployment, and use of AI systems. It includes
ethical considerations, accountability mechanisms, and compliance
with legal standards.

AI OFFICE: The European Commission’s function of contributing to the

implementation, monitoring and supervision of AI systems, general
purpose AI models and AI governance.

AI SYSTEM: AI system is a machine-based system designed to operate

with varying levels of autonomy and that may exhibit adaptiveness after
deployment and that, for explicit or implicit objectives, infers, from the
input it receives, how to generate outputs such as predictions, content,
recommendations, or decisions that can influence physical or virtual
environments (Art. 3 of the EU AI Act).

ALGORITHM: A set of sequential instructions applied on specific

input(s) to reach a predefined goal, ranging from the output of a basic
mathematical function to more complex tasks.

ARTIFICIAL INTELLIGENCE (AI): The development of computer systems

that can perform tasks that typically require human intelligence. These
tasks include learning, reasoning, problem-solving, perception, and
language understanding.

BIAS: Systematic and unfair preferences or prejudices in the data or

algorithms that can lead to discriminatory outcomes. Bias can arise
from the data used to train AI models or the design of the algorithms
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

themselves.

BIOMETRIC DATA: Personal data resulting from specific technical

processing relating to the physical, physiological or behavioural
characteristics of a natural person, such as facial images or
dactyloscopic data (Art. 3 of the EU AI Act).

CHATBOT: Software designed to mimic conversation with humans.

CLASSIFICATION: A type of machine learning task where the algorithm

assigns predefined categories or labels to input data.

CLOUD COMPUTING: The provision of computing services, such as

storage, computational power, and software applications, via the
internet.

CONFORMITY ASSESSMENT: The process of demonstrating whether the

requirements set out in the EU AI Act relating to a high-risk AI system
have been fulfilled.

COMPUTER VISION: A field of AI that enables machines to interpret and

make decisions based on visual data. Computer vision is used for image
and video analysis.

54
DEEPFAKES: AI generated or manipulated image, audio or video content
that resembles existing persons, objects, places or other entities or
events and would falsely appear to a person to be authentic or truthful
(Art. 3 of the EU AI Act).

DEEP LEARNING: Deep learning is a subset of machine learning and

AI that mimics human knowledge acquisition to enable models to
perform complex tasks like recognising patterns in various data types.
Encryption The process of converting information into a code to prevent
unauthorised access, particularly important for securing data in AI
applications.

EXPLAINABILITY: The ability to understand and interpret the decisions

and actions of AI systems, particularly in complex models like neural
networks.

FAIRNESS: Ensuring that AI systems treat all individuals and groups

fairly, without introducing biases or discrimination.

FOUNDATION MODEL: A pre-trained model that serves as the starting

point for various downstream tasks in machine learning.

GENERAL PURPOSE AI (GPAI): An AI system which is based on a general

purpose AI model that has the capability to serve a variety of purposes,
both for direct use as well as for integration in other AI systems (Art. 3
of the AIA)

GENERATIVE AI: AI systems that can generate new content, such as

images, text, or music, often using general purpose AI models.

HIGH-RISK AI: application AI applications with the potential for

significant impact on individuals’ rights and safety, requiring stricter
regulatory scrutiny.

LARGE LANGUAGE MODEL: A type of AI model, like GPT-4, that is trained

on vast amounts of text data and can generate human-like language.

MACHINE LEARNING: A subset of AI that focuses on the development

of algorithms that enable computers to learn and make predictions or
decisions based on data.

MARKET SURVEILLANCE AUTHORITY: A regulatory body responsible for

monitoring and ensuring compliance with market regulations, including
those related to AI applications.

NATURAL LANGUAGE PROCESSING: A subset of artificial intelligence,

focusing on how computers and human languages interact, allowing
machines to comprehend, decipher, and generate human language.

NEURAL NETWORKS: A set of algorithms, modelled loosely after the

human brain, designed to recognise patterns. Neural networks are a key
component of deep learning.

PROFILING: The process of analysing and classifying individuals based

on their characteristics, behaviours, or preferences.

QUANTUM COMPUTING: Quantum computing is a rapidly-emerging

technology that harnesses the laws of quantum mechanics to solve
problems too complex for classical computers.

55
 REGULATORY SANDBOX: A concrete and controlled framework set up by
a competent authority which offers providers or prospective providers
of AI systems the possibility to develop, train, validate and test, where
appropriate in real world conditions, an innovative AI system, pursuant
to a sandbox plan for a limited time under regulatory supervision (Art. 3
of the EU AI Act).

REMOTE BIOMETRIC IDENTIFICATION (RBI): system An AI system for the

purpose of identifying natural persons, without their active involvement,
typically at a distance through the comparison of a person’s biometric
data with the biometric data contained in a reference database (Art. 3 of
the EU AI Act).

ROBOTICS: The interdisciplinary field that combines computer science

and engineering to develop, build, and operate robots.

SOCIAL SCORING: The use of AI and data analytics to assess and score
individuals based on their behaviour, activities, or social interactions.

TRANSPARENCY: The openness and clarity in the functioning and

decision-making process of AI systems, ensuring that users and
stakeholders can understand the system’s behaviour.
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

56
Endnotes 1 Europol, 2021, Serious and Organised Crime Threat Assessment (SOCTA) 2021,
accessible at https://www.europol.europa.eu/publication-events/main-reports/europe-
an-union-serious-and-organised-crime-threat-assessment-socta-2021
2 European Parliament press release, June 2023, MEPs ready to negotiate first-ever rules
for safe and transparent AI, accessible at https://www.europarl.europa.eu/news/en/press-
room/20230609IPR96212/meps-ready-to-negotiate-first-ever-rules-for-safe-and-transpar-
ent-ai
3 European Parliament, 2023, Study to support the technical, legal and financial concep-
tualisation of a European Security Data Space for Innovation, accessible athttps://home-af-
fairs.ec.europa.eu/system/files/2023-02/Data%20spaces%20study_0.pdf
4 Y. J. Tan, S. Ramachandran, M. A. Jabar, et al., 2022, Financial Fraud Detection Based
on Machine Learning: A Systematic Literature Review, Applied Sciences, vol. 12, no. 19, pp.
9637, accessible at https://doi.org/10.3390/app12199637
5 K. Gülen, 2023, The parallel universe of computing: How multiple tasks happen simulta-
neously?, Dataconomy, accessible at https://dataconomy.com/2023/04/18/what-is-paral-
lel-processing/
6 Europol press release, 2020, Dismantling of an encrypted network sends shockwaves
through organised crime groups across Europe, accessible at https://www.europol.europa.
eu/media-press/newsroom/news/dismantling-of-encrypted-network-sends-shockwaves-
through-organised-crime-groups-across-europe
7 Europol press release, 2023, Dismantling encrypted criminal EncroChat communica-
tions leads to over 6 500 arrests and close to EUR 900 million seized, accessible at https://
www.europol.europa.eu/media-press/newsroom/news/dismantling-encrypted-criminal-en-
crochat-communications-leads-to-over-6-500-arrests-and-close-to-eur-900-million-seized
8 A. Babuta, 2017, Big Data and Policing: An Assessment of Law Enforcement Require-
ments, Expectations and Priorities, RUSI, accessible at https://static.rusi.org/201709_rusi_
big_data_and_policing_babuta_web.pdf
9 Ibid.
10 Ibid.
11 W. Hardyns, A.Rummens, 2018, Predictive Policing as a New Tool for Law Enforcement?
Recent Developments and Challenges, Eur J Crim Policy Res 24(2018), pp. 201–218, acces-
sible at https://doi.org/10.1007/s10610-017-9361-2
12 P. Walter et al., 2013, Predictive Policing: The Role of Crime Forecasting in Law En-
forcement Operations, RAND, accessible at https://www.rand.org/pubs/research_reports/
RR233.html
13 W.Hardynn & A. Rummens, 2017, Predictive Policing as a New Tool for Law Enforce-
ment? Recent Developments and Challenges, Eur J Crim Policy Res 24(2018), pp.201-2018,
accessible at https://doi.org/10.1007/s10610-017-9361-2
14 EUPCN, 2022, Artificial Intelligence and predictive policing: risks and challenges, acces-
sible at https://eucpn.org/sites/default/files/document/files/PP%20%282%29.pdf
15 Dutch Police, 2021, Crime Anticipation System, accessible at https://www.politie.nl/
wet-open-overheid/woo-verzoeken/landelijke-eenheid/woo-verzoeken-per-jaar/2021/
crime-anticipation-system.html
16 S. Oosterloo and G. van Schie, 2022, The Politics and Biases of the “Crime Anticipation ,
System” of the Dutch Police, accessible at https://ceur-ws.org/Vol-2103/paper_6.pdf
17 Europol, 2021, Internet Organised Crime Threat Assessment (IOCTA) 2021, accessible
at https://www.europol.europa.eu/publications-events/main-reports/internet-organ-
ised-crime-threat-assessment-iocta-2021
18 Europol, 2022, EU Terrorism Situation and Trend Report (TE-SAT) 2022, accessible at
https://www.europol.europa.eu/publication-events/main-reports/european-union-terror-
ism-situation-and-trend-report-2022-te-sat
19 Medium, 2023, Using OSINT to Improve Your Competitive Intelligence Strategy, acces-
sible at https://goldenowl.medium.com/using-osint-to-improve-your-competitive-intelli-
gence-strategy-eeb6114f9c71
20 Cambridge Consultants, 2019, Use of AI in online content moderation, accessible
at https://www.ofcom.org.uk/__data/assets/pdf_file/0028/157249/cambridge-consul-
tants-ai-content-moderation.pdf
21 European Commission, 2023, The impact of the Digital Services Act on digital plat-

57
 forms, accessible at https://digital-strategy.ec.europa.eu/en/policies/dsa-impact-plat-
forms
22 Meta, 2023, New Features and Additional Transparency Measures as the Digital Ser-
vices Act Comes Into Effect, accessible at https://about.fb.com/news/2023/08/new-fea-
tures-and-additional-transparency-measures-as-the-digital-services-act-comes-into-ef-
fect/
23 DeepLearning.AI, 2023, Natural Language Processing, accessible at https://www.
deeplearning.ai/resources/natural-language-processing/
24 P. Sarzaeim et al., 2023, A Systematic Review of Using Machine Learning and Natural
Language Processing in Smart Policing, Computers 2023; 12(12):255, accessible at https://
doi.org/10.3390/computers12120255
25 A. Dixon & D. Birks, 2021, Improving Policing with Natural Language Processing, In
Proceedings of the 1st Workshop on NLP for Positive Impact, pp. 115–124, accessible at
https://aclanthology.org/2021.nlp4posimpact-1.13.pdf
26 Ibid.
27 R. Abhijit, 2020, Understanding Automatic Text Summarization-1: Extractive Methods,
Towards Data Science, accessible at https://towardsdatascience.com/understanding-auto-
matic-text-summarization-1-extractive-methods-8eb512b21ecc
28 Roxanne project, NLP technologies against online crime, accessible at https://www.
roxanne-euproject.org/news/blog/nlp-technologies-against-online-crime
29 J. M. James et al., 2022, Digital Forensics AI: Evaluating, Standardizing and Optimizing
Digital Forensics Investigations, SpringerLink, accessible at https://link.springer.com/arti-
cle/10.1007/s13218-022-00763-9
30 H. Ravichandran, 2023, How AI Is Disrupting And Transforming The Cybersecu-
rity Landscape, Forbes, accessible at https://www.forbes.com/sites/forbestech-
council/2023/03/15/how-ai-is-disrupting-and-transforming-the-cybersecurity-land-
scape/?sh=5aab864c4683
31 J.S. Hollywood et al., 2018, Using Video Analytics and Sensor Fusion in Law Enforce-
ment, RAND, accessible at https://www.rand.org/pubs/research_reports/RR2619.html
32 Ibid.
33 A. Jain, K. Nandakumar & A. Ross, 2016, 50 Years of Biometric Research: Accomplish-
ments, Challenges, and Opportunities, Pattern Recognition Letters, 79, doi: 10.1016/j.
AI AND POLICING: THE BENEFITS AND CHALLENGES OF ARTIFICIAL INTELLIGENCE FOR LAW ENFORCEMENT

patrec.2015.12.013.
34 P. Grother et al.,2024, Face Recognition Technology Evaluation (FRTE). Part 2: Identifi-
cation, NIST, available at https://pages.nist.gov/frvt/reports/1N/frvt_1N_report.pdf
35 P. Grother et al.,2019, Face Recognition Vendor Test (FRVT) Part 3: Demographic
Effects, NIST, accessible at https://nvlpubs.nist.gov/nistpubs/ir/2019/nist.ir.8280.pdf
36 J. L. Mnookin, 2003, Fingerprints: Not a Gold Standard, Issues, XX (1) Fall, 2003, acces-
sible at https://issues.org/mnookin-fingerprints-evidence/
37 European Union Agency for the Operational Management of Large-Scale IT Systems
in the Area of Freedom, Security and Justice​(eu-LISA), 2020, The digital transformation
of internal security in the EU, AI and the role of eu-LISA, accessible at https://www.eulisa.
europa.eu/Newsroom/News/Pages/The-digital-transformation-of-internal-security-in-the-
EU-AI-and-the-role-of-eu-LISA.aspx
38 Y. Rawat, et al., 2023, The Role of Artificial Intelligence in Biometrics, 2023 2nd Inter-
national Conference on Edge Computing and Applications, pp. 622-626, doi: 10.1109/ICE-
CAA58104.2023.10212224.
39 A. Mitchell, 2013, Distinguishing Friend from Foe: Law and Policy in the Age of Bat-
tlefield Biometrics, Canadian Yearbook of international Law/Annuaire canadien de droit
international, 50, pp. 289-330, doi: 10.1017/S0069005800010869.
40 J. Thorpe, 2023, FBI amongst key adopters of innovative iris recognition technology for
law enforcement, International Security Journal, accessible at https://internationalsecurity-
journal.com/fbi-iris-recognition-law-enforcement/
41 J. Lunter, 2023, Synthetic data: a real route to eliminating bias in biometric, Biomet-
ric Technology Today 2023(1), accessible at https://www.magonlinelibrary.com/doi/

58
full/10.12968/S0969-4765%2823%2970001-5
42 A. Gomez-Alanis, J.A. Gonzalez-Lopez & A.M. Peinado, 2022, GANBA: Generative
Adversarial Network for Biometric Anti-Spoofing, Applied Sciences, 12(3):1454, accessible
at https://doi.org/10.3390/app12031454
43 Z.Zhang et al.,2023, An Improved GAN-based Depth Estimation Network for Face
Anti-Spoofing, ICCAI ‘23: Proceedings of the 2023 9th International Conference on
Computing and Artificial Intelligence, March 2023, pp. 323–328, accessible at https://doi.
org/10.1145/3594315.3594661
44 Europol, 2023, ChatGPT - the impact of Large Language Models on Law Enforcement,
A Tech Watch Flash Report from the Europol Innovation Lab, accessible at https://www.
europol.europa.eu/publications-events/publications/chatgpt-impact-of-large-language-
models-law-enforcement
45 P. Tomczak, 2018, Machine Learning and the Value of Historical Data, accessible at
https://kx.com/blog/machine-learning-and-the-value-of-historical-data/
46 C. Veal, M. Raper & P. Waters, 2023, The perils of feedback loops in machine learning:
predictive policing, Gilbert +Tobin, accessible at https://www.lexology.com/library/detail.
aspx?g=c8fff116-2112-48dd-841c-f9d1688d722b
47 N. Shahbazi, Y. Lin, A. Asudeh, & H. V. Jagadish, 2023, Representation Bias in Data:
A Survey on Identification and Resolution Techniques, ACM Computing Surveys (55)13,
accessible at https://doi.org/10.1145/3588433
48 EU Fundamental Rights Agency, 2022, Bias in algorithms - Artificial intelligence and
discrimination, doi:10.2811/25847
49 J. Burrell, 2016, How the machine ‘thinks’: Understanding opacity in ma-
chine learning algorithms, Big Data & Society, 3(1), accessible at https://doi.
org/10.1177/2053951715622512
50 S. Ali et al., 2023, Explainable Artificial Intelligence (XAI): What we know and what is
left to attain Trustworthy Artificial Intelligence, Information Fusion (99)2023, accessible at
https://doi.org/10.1016/j.inffus.2023.101805
51 B. Akhgar, et al., 2022, Accountability Principles for Artificial Intelligence (AP4AI) in the
Internal Security Domain AP4AI Framework Blueprint, accessible at https://www.ap4ai.eu/
sites/default/files/2022-03/AP4AI_Framework_Blueprint_22Feb2022.pdf
52 European Commission, 2019, Ethics guidelines for trustworthy AI, accessible at https://
digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai
53 European Commission, 2021, Proposal for a Regulation laying down harmonised rules
on artificial intelligence, accessible at https://digital-strategy.ec.europa.eu/en/library/pro-
posal-regulation-laying-down-harmonised-rules-artificial-intelligence
54 Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June
2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC)
No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and
(EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial
Intelligence Act)
55 L. Edwards, 2022, The EU AI Act: a summary of its significance and scope, Ada
Lovelace Institute, accessible at https://www.adalovelaceinstitute.org/wp-content/up-
loads/2022/04/Expert-explainer-The-EU-AI-Act-11-April-2022.pdf
56 Europol, 2023, The Second Quantum Revolution: the impact of quantum computing and
quantum technologies on law enforcement, accessible at https://www.europol.europa.eu/
publication-events/main-reports/second-quantum-revolution-impact-of-quantum-comput-
ing-and-quantum-technologies-law-enforcement#downloads
57 MIT Technology Review, 2023, AI-powered 6G networks will reshape digital interac-
tions, accessible at https://www.technologyreview.com/2023/10/26/1082028/ai-pow-
ered-6g-networks-will-reshape-digital-interactions/

59
About the Europol Innovation Lab
Technology has a major impact on the nature of crime. Criminals quickly integrate
new technologies into their modus operandi, or build brand-new business models
around them. At the same time, emerging technologies create opportunities for
law enforcement to counter these new criminal threats. Thanks to technological
innovation, law enforcement authorities can now access an increased number
of suitable tools to fight crime. When exploring these new tools, respect for
fundamental rights must remain a key consideration.

In October 2019, the Ministers of the Justice and Home Affairs Council called
for the creation of an Innovation Lab within Europol, which would develop a
centralised capability for strategic foresight on disruptive technologies to inform
EU policing strategies.

Strategic foresight and scenario methods offer a way to understand and prepare
for the potential impact of new technologies on law enforcement. The Europol
Innovation Lab’s Observatory function monitors technological developments
that are relevant for law enforcement and reports on the risks, threats and
opportunities of these emerging technologies. To date, the Europol Innovation
Lab has organised three strategic foresight activities with EU Member State law
enforcement agencies and other experts.

www.europol.europa.eu