Message to White House to take Cyber Security seriously - from MIT

KEEPING AMERICA SAFE:
TOWARD MORE SECURE
NETWORKS FOR CRITICAL
SECTORS

Cambridge, Massachusetts
March 2017

Joel Brenner
brennerj@mit.edu

Report on a Series of MIT Workshops, 2015-2016
With Recommendations for the New Administration

______

MIT Center for International Studies
https://cis.mit.edu

MIT Internet Policy Research Initiative
https://internetpolicy.mit.edu

______

MIT Center for International Studies 2 MIT Internet Policy Research Initiative

Table of Contents

Executive Summary and Recommendations 3
A History of Hesitancy 3
Background: 17
The Persistent Problem 17
Expanding Operational Risk 18
Why Are Systems Insecure? 20
Coordinating Research Policy 24
The Workshop Plan 25
The Sector-Specific Workshops 27
1. Electricity Sector Workshop 27
2. Financial Sector Workshop 33
3. Communications Sector Workshop 40
4. Oil-and-Natural-Gas Sector Workshop 45


Executive Summary and Recommendations
A History of Hesitancy

The digital systems that control critical infrastructure in the United States and most
other countries are easily penetrated and architecturally weak, and we have known it for a
long time. Yet Presidential leadership on infrastructure security has been hesitant and
chiefly rhetorical, while system operators have tended to focus on short-term fixes and
tactical improvements. Much effort has been devoted to developing better security
standards,1
but most standards are merely advisory. Key federal departments, notably but
not exclusively homeland security, defense, and energy have devoted significant effort to
improving infrastructure security. Examples would be too numerous to cite. But these
efforts have not altered the strategic balance.

Offense remains dominant. To break this cycle, the nation will require a
coordinated, multi-year effort to address deep strategic weaknesses in the architecture of
critical systems, in how those systems are operated, and in the devices that connect to
them. This effort must in part be technically directed, but it will also require a re-
evaluation of the laws, regulations, and policies that govern our networks. The challenges
we face are not merely technical. They are also economic, managerial, behavioral, political,
and legal. Indeed the technical challenges may be the easiest to address. For example,
aligning economic, tax, and liability incentives with the goal of higher security is not a
technical challenge. Re-aligning incentives would be a daunting task, but our critical
infrastructure cannot be made reasonably secure unless we do it.

This report identifies the most strategic of those challenges and proposes a policy
and research agenda that has the potential to achieve significantly higher levels of security
in critical networks over a five- to ten-year period. But the nation must begin now. Our
goal is action, both immediate and long-term.

To address this task, CIS and IPRI jointly convened a series of workshops focused on
four critical economic sectors, all of which are overwhelmingly or entirely in private hands:
electricity, finance, communications, and oil-and-natural gas (ONG). We did not set out to
write yet another description of the threat to our critical networks. In the wake of repeated,
widely reported foreign intrusions into our power grid and banking system and the recent
Russian interference in our national election, the threat is well known. Rather, we focused

1
See, e.g., National Institute for Standards and Technology, “NIST Releases Update to Cybersecurity
Framework,” January 10, 2017, at https://www.nist.gov/news-events/news/2017/01/nist-releases-
update-cybersecurity-framework, accessed February 9, 2017.


on what to do about it.

The workshops were attended by experts2
from leading enterprises in each sector,
by academic experts in relevant fields, and by a few government officials. We expected
commonalities across all four sectors, and we found many. Participants in each sector
bemoaned the difficulty of quantifying network risk, for example; and each workshop
expressed great interest in techniques of containing cascading failure. But we also
encountered differences among sectors – in part because the sectors operate in different
regulatory frameworks, and in part because two of these sectors – electricity and oil-and-
natural-gas (ONG) – are heavily dependent on industrial operating technology (OT) as well
as information technology (IT). Significant differences also exist within sectors as well as
between them in their levels of investment in cybersecurity and ability to fend off attacks.
We have preserved the essence of the individual workshops in summaries at the back of
this report.

The Recommendations

This report makes both long- and short-term recommendations of broad
applicability to critical infrastructure in the United States and, excepting certain legal and
regulatory matters, to critical infrastructure globally. The report identifies eight strategic
challenges to illuminate our predicament and guide our policy and research. Under each
challenge, it makes findings that emerged from the workshops and recommendations to
address them. The recommendations cover a wide range of issues, from the organization of
cybersecurity in the Executive Office of the President to technical measures of network
security and misaligned regulatory incentives. Each of the challenges is then followed by a
series of research questions whose answers could help meet that challenge. The report
therefore addresses three audiences: government officials, public and private institutions
that fund research, and the researchers themselves. By changing and focusing the research
environment, IPRI and CIS believe the nation could materially improve our long-term
security environment. We emphasize the coordination of funding, however; we do not
propose budgetary measures.

2
Participants were free to use any information received, but neither the identity nor the affiliation of
any speaker or participant could be revealed. Industry participants came from ten private energy
companies in the United States, Canada, France, and the United Kingdom, including two of the oil
majors; four leading international banks, a major data processor for financial institutions, and a leading
securities clearing organization; two tier-one communications providers; a leading computer chip
manufacturer; a leading maker of commercial and consumer software; and representatives of the
Government of Canada, the U.S. departments of homeland security and energy, and the Office of the
Governor of Massachusetts. Participants from firms and governments in India and from another U.S.
university were invited but did not attend. The views expressed in this report do not necessarily reflect
those of individual workshop participants or of their enterprises and agencies.


Some of these research questions we pose are broad and technical (e.g., Can cyber
risk be measured?); others are narrow and focus on non-technical impediments to adopting
technically available security measures (e.g., What economic or other factors impede the
adoption of secure connections between service providers?). Differences in generality were
unavoidable if we were to describe the full range of technical and policy questions that
must be answered, especially because many of the impediments are legal, economic, and
political rather than technical. Taken together, these questions should form the basis of a
focused, national agenda that must be adopted, coordinated, and funded if we are to
escape from a twenty-five-year cycle of futile tactical measures and imprecise aspirational
statements from a never-ending series of governmental and private groups.

The nation can no longer afford a pattern of uncoordinated executive action and
scattershot research. Total security is not achievable. But a materially improved security
environment for the infrastructure on which virtually all economic and social activity
depend can be created with sufficient resources and political will. Achieving this goal will
require a more determined and more directive approach from the highest levels of
government and industry. It will also require more energetic and coordinated steps from
the President than any of his predecessors has been willing to take.


FINDINGS AND RECOMMENDATIONS
FIRST CHALLENGE
Improve Coordination.
Finding:

Critical infrastructure defense is insufficiently coordinated across the
government. Changing the status quo will require a more directive effort
from the White House.

Recommendation:

The President should elevate his cybersecurity advisor to the position of
deputy national security advisor for cybersecurity. That official should be
directed and empowered to work with the Office of Management and
Budget (OMB) to focus long-term policy across the government on the
substantive challenges identified below and to produce on an accelerated
schedule a federal research agenda and budget for the cybersecurity of
critical infrastructure focused on these same challenges. OMB should
determine that funds are spent accordingly.

SECOND CHALLENGE
Measure cyber risk and infrastructure fragility.
Finding:

Quantifying risk in either absolute or relative terms is a difficult challenge
that impedes cybersecurity investment in all sectors examined except
certain financial institutions. The asserted inability to measure the rate of
return on cybersecurity investment is a closely related problem3
that
affects overall investment levels and makes it difficult to target investment.
Fragility of systems is a salient aspect of risk that concerned participants in
all sectors. Absent assurances of confidentiality, candid participation by the
private sector will not occur. However, the public should be informed of
the general state of security of critical infrastructure.

3
Most participants accepted the view that cyber risk, changes in cyber risk resulting from a specific
security investment, and the rate of return on that kind of investment could not be measured. For the
contrary view, see Douglas W. Hubbard and Richard Seiersen, How to Measure Anything in
Cybersecurity Risk (New York, 2016).


Recommendation:

The President should direct the lead departmental secretary to convene
on an accelerated schedule a meeting of representatives of the relevant
national laboratories and other experts to assess impediments to
measuring cyber risk and fragility and to recommend a national strategy
to meet this challenge. The meeting should be closed to the public and its
proceedings, though not the strategy, should be kept confidential.

Research Questions:

1. Can cyber risk or network fragility be measured? Can changes in risk as the result
of specific security investments be measured? If so, why are enterprises not doing
it?

2. Would the answers to these questions produce more rational decision-making by
enterprises? If not, why not?

3. Can simulation-based modeling be used to create cybersecurity stress-tests for
critical sectors? In the electricity sector, could that type of modeling be used to
test the ability to “cold start” electricity generation? Can the results of such
modeling be protected from public disclosure? How, and at what level of
generality, should the public be informed of vulnerabilities in critical systems?

4. Should the answers to these questions have regulatory implications for some or all
critical sectors?

5. Can the necessary de-identified4
data be obtained to support research into these
questions? Would legislation be appropriate to compel the production of that de-
identified data in the interest of national security – but with an exemption from
disclosure and under a legal privilege that would prevent its use for any other
purpose?5
How would the required data be defined, and who should hold it?

4
De-identification means removing identifying aspects of data so that, practically speaking, it would be
difficult and expensive to re-associate it with a particular person. Perfect anonymization of data is not
possible in most circumstances.
5
The National Infrastructure Protection Act, codified as 42 U.S.C. §§ 5195c et seq., does not clearly
give the Department of Homeland Security power to require production of specific categories of data
from private firms. See 42 U.S.C. § 5195c (d)(2)(A) and (B).


THIRD CHALLENGE
Review laws and regulations with the goals of reducing risk and optimizing security
investment.

Finding:

Participants from all sectors overwhelmingly believed there was a material
disconnection between mandatory compliance regimes and improvements
in cybersecurity. Most participants from all sectors except finance believed
that federal tax and regulatory incentives for higher levels of cybersecurity
investment should be considered. Many participants from the electricity
and telecommunications sectors believed that regulations either impeded
or did not encourage higher levels of cybersecurity investment.

Recommendations:

The President should propose legislation at the earliest opportunity for
the more favorable tax treatment of qualified cybersecurity investment in
critical infrastructure and, potentially, throughout the economy, including
investment necessary to convert to a more secure DNS and to more
secure border gateway protocols. To qualify for favorable treatment,
investments should be in products and services that are demonstrably
compliant with the framework promulgated by the National Institute for
Standards and Technology (NIST).6

The secretary of energy, state public utility commissioners, and the
National Association of Regulated Utility Commissioners should forthwith
examine the effect of utility regulation on cybersecurity with particular
attention to (i) the effect of current regulations on cybersecurity
investment and (ii) the usefulness of current compliance standards in
achieving higher levels of security.

6
National Institute for Standards and Technology, “Framework for Improving Critical Infrastructure
Cybersecurity,” version 1.0, February 12, 2014, at
https://www.nist.gov/sites/default/files/documents/cyberframework/cybersecurity-framework-
021214.pdf, accessed February 20, 2017. For draft version 1.1 of the Framework, see
021214.pdf, accessed February 20, 2017.


Research Questions:

1. How should liability rules and regulations be optimized to produce more secure
behavior by vendors and by the owners and operators of infrastructure? A
comparative study of state as well as federal regulatory models would be useful in
addressing this question.

2. Can cybersecurity regulation be harmonized across government? Government
regulates by sector. For example, the Federal Communications Commission
regulates telecommunications; the Treasury Department, the Federal Reserve, and
other agencies regulate banks; the Energy Department, the Environmental
Protection Agency, and the states regulate energy, and so on. But as these
“vertical” regulators have also begun to regulate cybersecurity, a complex of
overlapping, expensive, and potentially inconsistent standards is emerging. Are
these regulations driving greater security, or are they merely more elaborate and
expensive compliance regimes?

3. The many competing compliance standards create confusion. Should the
government make the NIST Framework, and only the NIST Framework, a single
mandatory standard across government and for contractors dealing with the
government?

4. Could the financial impact on insurers and re-insurers of the damage resulting
from a successful attack on one or more critical sectors be absorbed by them? If
not, what law and policy would be required to make it likely that such losses could
be absorbed?

5. Can the necessary, de-identified data be obtained to support research into these
questions? If not, would legislation be appropriate to compel the production of
that de-identified data in any sector while protecting the rights of the enterprises
that would produce the data? How would the required data be defined?

FOURTH CHALLENGE
Enable critical infrastructure operators to quickly identify and respond to
cyber risk arising from cross-sector linkages as well as from their own networks.

Finding:

All sectors depend on electricity, and the financial sector’s global platform
supports transactions with energy and telecommunications. These and
other linkages create possibilities for cascading failure that are
insufficiently understood and not adequately illuminated by sector-specific


simulations and other testing. Participants from all sectors stated
emphatically that cooperation on pooling proprietary data and candor
regarding the results of testing could not be achieved unless parties could
be assured that the data and results would remain confidential and could
not be used for other purposes.

Recommendation:

The President should direct the lead departmental secretary to convene
on an accelerated schedule a meeting of representatives of the nation’s
leading industrial insurers and other experts to examine the steps
necessary to enable more robust cross-sector simulations, including the
sharing of data, and to make appropriate recommendations to the
President. The meeting should be closed to the public and its proceedings
kept confidential, but the resulting recommendations should be public.

Research Questions:

1. What steps would increase the likelihood of early detection of a slow-moving
strategic attack on a critical sector or across critical sectors? How will detection
techniques be affected by the anticipated move to IPv6?7

2. How would such an attack affect critical backup systems?

3. Can simulation-based modeling be used to create better cross-sector stress tests?

4. Can simulated cyber disasters help determine how communications should be
prioritized in the event of a national emergency?

5. Can efforts to use big data and fast processing to quickly detect intrusions in
critical networks be accelerated?

6. What, if anything, prevents the effective use of identity management tools across
the full range of steps necessary to execute a successful exploit or attack?

7
IPv6 is an Internet addressing protocol that would expand the number of IP addresses available under
the current protocol, called IPv4, by a factor of 7.9 x 1028
. It could therefore render ineffective current
techniques for scanning systems for malware because the address space to be scanned would be
exponentially larger.


FIFTH CHALLENGE
Reduce component complexity and the vulnerabilities inherent in them.

Finding:

Participants from the electricity and energy and oil-and-natural gas (ONG)
sectors believed that unduly complex, and insufficiently secure, hardware,
software, and industrial controls were a significant source of cyber
vulnerabilities that created physical danger as well as risk to information.
Participants from the ONG sector were emphatic on this point. Both energy
sectors are highly dependent on industrial operating technology. This is a
significant supply chain risk created by commercial, not technological,
factors. Suppliers find it profitable to market cheap, general purpose
hardware and software for multiple uses, regardless of differing security
tolerations in different sectors and uses.

Recommendation:

The President should direct the lead departmental secretary to report to
him on an accelerated schedule on the feasibility, timeline, and expense
of supporting and otherwise incentivizing the production and use of more
secure and less complex hardware, software, and controls for use in
critical infrastructure.

Research Questions:

1. Can the technical, economic, and regulatory obstacles to reducing complexity in
both information technology and industrial operating technology be identified?

2. Field programmable gate arrays (multipurpose computer chips) are cheap, so they
are used for many purposes including commercial routers and industrial controls
used in critical infrastructure operations, but their complexity and superfluous
functionality increase risk. The same may be said of general purpose processing
units, operating systems, and software systems.

a. Can standards be established to reduce the vulnerabilities in logic processors
and the software and firmware that control them?

b. Can standards be established, or incentives created, to phase out design tools
that permit hardware and software designers to make the same basic errors
repeatedly, such as allowing buffer overflows?


c. What steps would be necessary to establish a certification system for hardware
and software, possibly modeled on the Underwriters Laboratory for electrical
products?

d. Can microchips be designed so that entire sectors of those chips can be
cheaply, reliably, and verifiably disabled so that functionality matches task
requirements?

3. What incentives should be in place to induce controls manufacturers and Internet
service providers to use less vulnerable chips?

4. Are the departments of defense, energy, and homeland security optimizing their
role in creating and supporting a market for simpler and more secure commercial
devices in critical infrastructure? For example, can these departments jointly
establish metrics for complexity and standards for controls, and use their
procurement decisions to favor less complex and more secure hardware and
software?

5. Can simpler firmware and operating systems be cost-effectively developed and
marketed for use in critical infrastructure?

Sixth Challenge
Address fundamental issues of system architecture.
Findings:
1. The Internet is a legacy system designed for non-commercial uses with
little or no need for security. Security has chiefly been an option for end
points, which frequently ignore it in favor of speed-to-market and low
costs. Hardware and software that run on the Internet display wide
differences in security, and the tools for creating hardware and software
enable many of the same security errors to be repeated over many years,
without liability.
2. Security professionals from all sectors overwhelmingly believed that
certain aspects of their systems could not otherwise be made reasonably
secure unless isolated from public networks. There are significant
differences of opinion about appropriate degrees of isolation.


Recommendations:
1. The President should direct the secretaries of energy and homeland
security:

a. in consultation with the Federal Energy Regulatory Commission (FERC), to
explore the feasibility, expense, and timelines of isolating from public
networks8
all controls and operations of activities within FERC’s
jurisdiction,9
to define acceptable degrees of isolation, and to report to
the President on an accelerated schedule; an

b. in coordination with the FERC and the North American Electric Reliability
Corporation (NERC),10
to convene at the earliest practical time a
conference of state electricity regulators to explore the feasibility and
expense of isolating key elements of electricity generation and delivery
from public networks.

2. The President should direct the lead departmental secretary to consult
with key stakeholders, including vendors, users, the public, and the
insurance industry, about the desirability and feasibility of (i)
establishing legally binding standards of care in the manufacture of
hardware and software for critical infrastructure, and (ii) the
establishment of a privately owned and managed accreditation bureau
for such hardware and software, and to report to the President on an
accelerated schedule.

Research Questions:
1. Should some operations of some or all critical sectors be isolated from the
Internet? If so, which ones? How should “isolation” be defined? What level of
isolation would be appropriate for particular systems in critical applications? Who
should determine that?

8
This is not a recommendation to create a single non-public energy network. Isolation from public
networks does not imply isolation from efficient, digital operating systems that produce real-time, or
near real-time, information about those systems. Non-public information and operating systems based
on TCP/IP protocols are available or can be created.
9
FERC has jurisdiction over the interstate transmission of electric power. Power generation and
delivery are regulated by the states and territories.
10
NERC is composed of the owners and operators of the grid and has been named by FERC as the
“Electric Reliability Organization.” It is charged by Congress to “establish and enforce reliability
standards for the bulk-power system,” subject to FERC’s oversight.


2. Can block chain or other technology be used to verify accounts in a timely fashion
to reduce the risk of corrupted backup systems and wiped accounts?

3. What changes to security architectures would let us more efficiently manage
system accesses and identities for devices, people, applications, and data, both
internally and externally?

4. Can a system be designed so that its failure would be immediately transparent to
its operator? Can the state of the system’s algorithms be made understandable to
humans? Would it be cost-effective to impose audit requirements on that kind of
system? (E.g., if a driverless car ran off a bridge, could its control algorithm be
made to explain why it did that?) If so, why don’t we mandate that kind of
auditability in critical sectors?

5. What economic, regulatory, or other factors impede the more rapid phasing out of
legacy components of electronic systems in favor of components that are not
merely newer but are demonstrably more secure?

6. What economic or other factors impede the adoption in the private sector of the
existing but largely unused secure domain name system or an alternative security
architecture? What incentives could accelerate the adoption of a more secure
domain name system?

7. In the communications sector, what economic or other factors impede the
adoption of secure border gateway protocols that would make it impossible, or
substantially more difficult, to divert network traffic? What incentives could
accelerate the adoption of that type of control?

8. Companies have differing interests. Academics make a living by disagreeing with
one another and often prefer the notional perfect to the achievable good. Universal
agreement on a domain name system and border gateway controls is therefore not
achievable. Is there a point, short of war, when the Congress should make these
choices?

9. The Internet of Things makes attack surface management geometrically more
difficult. What aspects of insecure devices matter most in this respect? Should
enhanced security be applied at the device level or only at higher levels within
networks?

10. Would it be feasible and efficient in a virtual network to segregate or at least
identify all executable code, thus making unauthorized executables more readily
discoverable?


SEVENTH CHALLENGE

Formulate an effective deterrence strategy for the nation.

Findings:

The cybersecurity postures and capabilities of the United States and its
peer or near-peer competitors in cyberspace have served to deter outright
attacks against one another’s critical infrastructure, but have been
unsuccessful in deterring lower-level but increasingly harmful cyber
operations across our economy, society, and political system. Hostile acts
are systematically carried out below the level of armed conflict that have
the potential to gradually reduce this nation’s stature and security and its
ability to lead free and open democracies around the globe. In this gray
space between war and peace, the United States does not have an
effective deterrence strategy against either nation-states or transnational
groups bent on terror or other forms of disruption of our critical
infrastructure.

Recommendation:

The President should direct his national security advisor to review the nation’s
deterrence strategy. That strategy should include, but not be limited to, (i)
hardening critical American systems and infrastructure; (ii) raising the price
for attacking them; (iii) constructing a diplomatic strategy for achieving
verifiable cybersecurity agreements with potential adversaries; and (iv)
evaluating the nation’s ability in the long term to maintain offensive
dominance in cyberspace and the stabilizing or destabilizing effect of
attempting to do so.

Research Questions:
1. In view of the demonstrated ability of certain nation-states to exploit critical
networks for economic, political, and potentially military advantage, would a more
directive policy toward hardening critical networks be justified? Would that course
of action be politically acceptable in the United States and among other nations
involved in global transactions and telecommunications?

2. Cyber network operations by capable nation-states and their proxies are difficult or
impossible to prevent, yet we expect critical infrastructure operators to defend
themselves against these attacks. Is this the right public policy? If not, what policy
should replace it?


3. Will the pursuit of offensive dominance in cyberspace continue to be feasible in the
next five to ten years? Will its pursuit be inconsistent with order and stability in
cyberspace, as it proved to be in the strategic nuclear relationship with the Soviet
Union? What are the implications of the answers to these questions for American
diplomatic strategy in cyberspace?

Is the President receiving robust counter-strike options, both military and non-
military, for cyber intrusions, including those that do not rise to the level of
armed conflict under international law?

4. Is any department of government conducting realistic simulations and other
exercises to explore the consequences of non-military counter-strikes in response to
a cyberattack? Does the President’s understand and approve of the assumptions
that underlie these exercises?

EIGHTH CHALLENGE
Accelerate and improve the training of cybersecurity professionals.

Findings:

There is a serious dearth of cybersecurity expertise in the United States,
especially at advanced levels. The nation does not produce enough
graduates with advanced cybersecurity skills or with skills in both
cybersecurity and in the operation of industrial operating systems.

Recommendation:
The President should appoint a blue-ribbon commission on the
feasibility of increasing the supply of highly trained computer scientists
and engineers and developing model curricula for training computer
scientists and engineers in the defense of critical systems. The
commission should report to the President within 180 days.


This is a time for action. It is also a time for calm, long-term strategic
thinking, based on sound research, into the underlying causes of cyber
insecurity and how to address them.

Research Questions:
1. Adm. Hyman Rickover created a rigorous model for selecting and training nuclear
submariners. Should government or industry adopt his model for the cyber
defense of critical infrastructure?

2. Can effective network defense skills be taught without also teaching high-level
offensive skills? If not, given the risk of teaching those skills to a wider cadre, who
should be eligible to receive that instruction? Should qualified trainers, in defined
circumstances, be granted liability protection for teaching offensive tactics?

3. Are different core curricula appropriate to train people to operate and defend the
networks of different critical infrastructures? If so, who should develop them?

4. Should people in cybersecurity disciplines be subject to specialized training and
certifications, as in other professional disciplines?

Background:
The Persistent Problem

In the United States, Presidential Directives to address infrastructure risk have
emerged from the White House like clockwork for more than twenty-five years. In 1990,
President George H.W. Bush announced to the country what intelligence officials, but not
many others, already understood: “Telecommunications and information processing
systems are highly susceptible to interception, unauthorized electronic access, and
related forms of technical exploitation, as well as other dimensions of the foreign
intelligence threat. . . . “

In 1998, as enterprises were beginning to shift both information systems and
operations to the Internet, President Clinton warned of the insecurities created by cyber-
based systems. In 1998 he directed that “no later than five years from today the United
States shall have achieved and shall maintain the ability to protect the nation’s critical
infrastructures from intentional acts that would significantly diminish” our security. Five
years later would have been 2003.


In 2003, President George W. Bush implicitly recognized that this goal had not
been met. He stated that his cybersecurity objectives were to “[p]revent cyber attacks
against America’s critical infrastructure; [r]educe national vulnerability to cyber attacks;
and [m]inimize damage and recovery time from cyber attacks that do occur.” Meanwhile,
virtually all commercial and operational activity was migrating to the Internet, which
remained insecure.

By 2009, concerns about critical infrastructure had become acute. President
Obama said:

The architecture of the Nation’s digital infrastructure, based
largely on the Internet, is not secure or resilient. Without major
advances in the security of these systems or significant change in
how they are constructed or operated, it is doubtful that the
United States can protect itself from the growing threat ….

By 2013 – fifteen years after President Clinton had said the country’s critical
infrastructure should be secure from malicious disruption by 2003 -- President Obama
acknowledged that the goal had not been met: “The cyber threat to critical infrastructure
continues to grow and represents one of the most serious national security challenges we
must confront.” The view at the enterprise level is much the same. Trend Micro, a leading
Internet security firm, reported in 2015 that critical infrastructure operators throughout
the Western hemisphere “painted a picture that depicts the threat [to their networks] as
being severe, while some perceived the future of securing these infrastructures as bleak.”

The question the nation faces is therefore this: Are we condemned to remain in
this unstable and insecure condition, in which the best we can do is to repeat urgent but
futile warnings from high places and, at the operational level, merely to refine our tactics
in a losing game of Whac-A-Mole? To find an answer, we gathered experts from industry,
government, and academia, to imagine – in President Obama’s phrase – “a significant
change in how [systems] are constructed or operated.” This meant going beyond the
intense and difficult day-to-day tactical challenges that critical sector operators face,
important as they are, to imagine a better security environment in five to ten years and to
understand what keeps us from getting there.
Expanding Operational Risk
For the owners and operators of critical infrastructure, the prime concern is risk to
continuity of operations rather than theft of information, though that, too, is a serious
risk. An intruder who can steal massive amounts of data from a system remotely can also
corrupt the information on the system, or wipe information from it, or shut it down.


Information technology and industrial operating technology have largely converged. A
decade ago, researchers at the Idaho National Laboratory proved they could physically
destroy a diesel-electric generator using only a keyboard and a mouse.11
Real-world
examples soon followed.

In 2010, the centrifuges used to enrich uranium gas at Iran’s Natanz nuclear facility
started failing rapidly. The Iranians were baffled – until researchers in Germany diagnosed
the Stuxnet virus, now widely attributed to the intelligence services of the United States
and Israel.12
In 2012, cyber attacks from Iran wiped all information from thirty thousand
computers at the world’s largest oil refiner, Saudi Aramco.13
In 2014, an unidentified
intruder used a spear-phishing ruse to gain access to the network of a German steel mill,
then caused multiple components of the industrial control system to fail, resulting in
massive physical damage.14
Meanwhile, starting in 2011, a Russian operation known as
“Dragonfly/Energetic Bear” began targeting North American aviation companies before
shifting to U.S. and European energy firms. Its targets included “energy grid operators,
major electricity generation firms, petroleum pipeline operators, and Energy industry
industrial control system (ICS) equipment manufacturers. Most of the victims were in the
United States, Spain, France, Italy, Germany, Turkey, and Poland.”15
There were no
reports of damage from these penetrations; they appeared to be reconnoitering exercises
that could facilitate damaging attacks on the systems later, if the intruder chose to attack.
In 2015 the prospect that an attacker might launch a damaging attack on an adversary’s
energy grid became reality when portions of Ukraine’s power grid were disabled for
several hours in a coordinated attack on three energy firms. This was the first publicly
acknowledged attack on a power grid. The Ukraine government immediately blamed
Russia. The attackers employed a range of sophisticated tools, but in the view of several
analysts, “the strongest capability of the attackers was not in their choice of tools or in
their expertise, but in their capability to perform long-term reconnaissance operations
required to learn the environment and execute a highly synchronized, multistage,
multisite attack.”16

11
“The experiment used a computer program to rapidly open and close a diesel generator's circuit
breakers out of phase from the rest of the grid and cause it to explode.” Wikipedia, “Aurora Generator
Test,” at https://en.wikipedia.org/wiki/Aurora_Generator_Test, accessed January 6, 2017.
12
Wikipedia, “Stuxnet,” at https://en.wikipedia.org/wiki/Stuxnet, accessed November 16, 2016.
13
Nicole Perlroth, “In Cyberattack on Saudi Firm, U.S. Sees Iran Firing Back,” New York Times, October
23, 2012, at http://www.nytimes.com/2012/10/24/business/global/cyberattack-on-saudi-oil-firm-
disquiets-us.html, accessed December 23, 2016.
14
R.M. Lee et al., “German Steel Mill Attack,” SANS Institute, ICS Defense Use Case, December 30,
2014, at https://ics.sans.org/media/ICS-CPPE-case-Study-2-German-Steelworks_Facility.pdf, accessed
December 23, 2016.
15
June 30, 2014, at https://www.symantec.com/connect/blogs/dragonfly-western-energy-companies-
under-sabotage-threat, accessed December 6, 2016.
16
Lee et al., “Analysis of the Cyberattck on the Ukrainian Power Grid,” SANS Institute, ICS Defense Use
Case, March 18, 2016, at http://www.nerc.com/pa/CI/ESISAC/Documents/E-
ISAC_SANS_Ukraine_DUC_18Mar2016.pdf, accessed December 23, 2016.


This is the offense-dominant environment that critical infrastructure operators
now live in. Network defense has certainly gotten better in the last fifteen years in
absolute terms, but so has the offense. Relative to the increased resources and
sophistication of criminal and nation-state attackers, it is doubtful the defense has
improved at all. Attacks are still easy and cheap to launch and difficult and expensive to
defend against.

The offense continues to enjoy inherent advantages owing to human fallibility,
architectural flaws in the Internet and the devices connected to it, massive data
aggregation, and pervasive interconnectivity. And the attacker must succeed only once,
while the defense must succeed thousands or millions of times. Connecting geographically
dispersed operating equipment to the Internet has brought undoubted efficiencies to
electricity generators and other industries, but it has also created dangerous
vulnerabilities in the systems that keep the lights on and power the economy. In late
2016, the recently retired chief security officer of AT&T said it was “inevitable that
significant, large-scale cyber attacks will be launched against our critical infrastructure [in
the coming four years]. These attacks will shift from the theft of intellectual property to
destructive attacks aimed at disrupting our ability to live as free American citizens. I do
not know of a single cyber security expert in our country who would disagree with this
view.”17
We concur.
Why Are Systems Insecure?
When the Internet was being designed in the early 1970s, it was not initially clear
what the important security issues were. Its initial purposes were to assure
communications in the event of a nuclear attack through packet-switched routing, and
then to serve as the basis for collaboration among geographically dispersed scientists
working for the Department of Defense. The relatively few people having access to the
original network were a trusted group for whom security was not an issue. Insofar as the
network’s sponsors in the Department of Defense and the intelligence community
thought about security, they preferred that security challenges be pushed onto the
attached end-nodes, without appreciating the difficulty of doing so. The Internet’s
designers understood that many security problems would best be addressed through
encryption, but encryption was not a commercially practical technology at the time for
reasons of performance and lack of open standards. At the time, encryption was also
regulated as a munition for export purposes. These considerations, together with the
imperative to get the Internet to work at all, led to several classes of security problems. In
particular:

17
Edward Amoroso, “An Open Letter to the President-Elect on Cyber Security,” LinkedIn, November
25, 2016, at https://www.linkedin.com/pulse/open-letter-president-elect-cyber-security-edward-
amoroso, accessed December 10, 2016.


1. Several of the core control protocols and supporting services of the Internet were
designed without an approach to security, and adding security after the fact has
proved more difficult than anticipated. These protocols include the global, inter-
domain routing protocol (Border Gateway Protocol or BGP), the Domain Name
System (DNS),18
and the Certificate Authority system. In all these cases, secure
alternatives have been proposed but have not been taken up in the marketplace.
What the original designers thought would be a technical challenge has turned out
in all cases to be a challenge created by misaligned economic incentives, poor
coordination and leadership, a lack of global trust among stakeholders, and
disagreements about what the security problems are.

2. Strictly speaking, the Internet is simply the network that connects end-points using
a technical protocol called “TCP/IP.”19
It was never meant to police itself for
criminal or offensive behavior. To a significant degree, therefore, the Internet is
doing what it was designed to do: that is, to connect end-points. Many (perhaps
most) of the vulnerabilities in our systems occur at other levels – in hardware
designed with little or no consideration for basic security, for example;20
in
carelessly written software;21
and in applications created for quick market
penetration that are unable to meet reasonable security requirements.22
In the
early days of the Internet’s development, the designers paid relatively little
attention to the challenge of developing secure applications, since in their view
they had no control over what application designers could do. Most application
designers today are motivated by features, time to market, and return on
investment. These priorities align poorly with security. This set of actors is highly
diverse, unregulated, transnational, and sometimes hard to find, and it is not clear
what approach could be used to nudge them to attend more to security.

18
“Domain Name System,” Wikipedia, at https://en.wikipedia.org/wiki/Domain_Name_System,
accessed December 12, 2016.
19
For definitions of the Internet and TCP/IP protocols, see respectively Wikipedia at “Internet,”
https://en.wikipedia.org/wiki/Internet, and “Internet Protocol Suite,”
https://en.wikipedia.org/wiki/Internet_protocol_suite, both accessed January 7, 2017.
20
For the IoT attack on an important Internet company, see Schneier on Security blog, “Lessons from
the Dynamics’s DDoS Attack,”
https://www.schneier.com/blog/archives/2016/11/lessons_from_th_5.html; James Scott and Drew
Spaniel, Rise of the Machines: The Dynamics’s Attack Was Just a Practice Run, December 2016,
Institute for Critical Infrastructure Technology report, at http://icitech.org/wp-
content/uploads/2016/12/ICIT-Brief-Rise-of-the-Machines.pdf, accessed January 8, 2017.
21
See, e.g., Wikipedia, “Buffer Overflows,” at https://en.wikipedia.org/wiki/Buffer_overflow, accessed
January 3, 2017. Buffer overflows have been known to be a security vulnerability for years.
22
See, e.g., Lucian Constantin, “App Developers Not Ready for Stricter IoS Security Requirements,”
Computerworld, December 6, 2016, at www.computerworld.com/article/3147373/security/app-
developers-not-ready-for-stricter-ios-security-requirements.html, accessed December 7, 2016.


3. There is no agreement today on who, if anyone, should be responsible for making
the Internet ecosystem more secure. For example, it can be extremely difficult,
even impossible, to be certain who you are communicating with on the Internet.
Identities can be easily spoofed and websites counterfeited, enabling fraud. But
which actors in the Internet ecosystem should undertake to fix this? Should the
packet-forwarding layer of the Internet attempt to impose a single, global identity
scheme that applies to all applications? Doing so would raise yet again the
question of global trust and coordination. It would make anonymous action very
difficult. That would reduce crime, but it would also enhance surveillance powers
and thereby threaten privacy. Should the large and uncoordinated community of
application designers be told that identity assurance is their problem? In fact, the
solution probably requires support at all layers. But there is no institutional forum
in which an allocation of responsibility can be resolved.

4. Data files, which are passive, and executable files, which perform operations on
data, cannot be distinguished as they are transmitted across the Internet. But this
approach left the discrimination between data and executable files to the
application designers in the end-nodes, who were often indifferent to the issue. As
a result, malicious executables are easily disguised among large quantities of data.
They are easy to insert and extremely difficult to find in a large database or
system. This problem became much more difficult once data files (e.g., a Word
file) were designed to embed executable code (e.g., macros).

After Congress made the Internet generally available for commercial use in 1992,
the network became the backbone of our entire system of economic and social
communication, and increasingly of our physical operations, so these inherent
weaknesses assumed enormous significance. As Richard Danzig has noted, “Cyber systems
create serious security problems because they concentrate information and control and
because the complexity, communicative power and interactive capabilities that enable
them unavoidably create vulnerabilities.”23
Putting massive amounts of information in
one place, which is highly efficient, also facilitates massively efficient theft. And
connecting almost everything to almost everything else, which is also efficient, means
that a vulnerability in any part of the interconnected system is a vulnerability in every part
of it. These factors, together with the difficulty of tracing and attributing attacks, make
the Internet a prime environment for criminals.

23
Richard Danzig, “Surviving on a Diet of Poisoned Fruit: Reducing the National Security Risks of
America’s Cyber Dependencies,” Center for a New American Security (July 2014), p. 9, at
https://www.cnas.org/publications/reports/surviving-on-a-diet-of-poisoned-fruit-reducing-the-
national-security-risks-of-americas-cyber-dependencies, accessed December 24, 2016.


It is a serious error to assume that cybersecurity is entirely a matter of technical
specifications and system design. Poor business management, lack of clear responsibility
within organizations, and bad user behavior would continue to create significant
vulnerabilities even if the technical issues could suddenly be fixed. Last year, when for the
first time the Bank of England included cybersecurity as a major risk factor for the
financial stability of the United Kingdom, its number one finding was, “Overemphasis on
technological (as opposed to management, behavioural and cultural) aspects weakens
cyber defensive capabilities.”24
We concur.
A common human error enabling fraud is susceptibility to an online scam known
as phishing. Phishing involves sending a mass email that appears to come from a trusted
source such as a bank or a well-known company, but does not. A recipient (the “phish”)
who opens the email and clicks on the attachment unwittingly downloads malware. The
purpose of the malware varies. It may steal information such as passwords or credentials,
or it may enlist the recipient’s machine in a campaign to advertise pornography, drugs,
etc. Phishing campaigns are nearly cost-free to conduct and are highly successful.
According to Verizon, thirty percent of recipients open phishing emails, and about a third
of them click on the attachment. “The median time for the first user of a phishing
campaign to open the malicious email [was] 1 minute, 40 seconds. The median time to
the first click on the attachment was 3 minutes, 45 seconds ….”25

Spear phishing is a socially engineered fraud aimed at a specific person, often a
corporate or government official. This is a favorite tactic of sophisticated criminal gangs
and intelligence services, which can craft an email that appears to come from a trusted
person on a topic that the recipient is known to be interested in. Sometimes the malware
is automatically downloaded merely by opening the email. In a recent survey by Trend
Micro, “spear-phishing tactics were cited by all responding members as the single biggest
attack method they had to defend against, with the exploitation of unpatched vendor
software vulnerabilities being a distant second.” Whether an effective technological
defense to this vulnerability can be deployed remains to be seen.

Weaknesses in the email system also contribute to identity spoofing. The basic
design of email is older than the Internet; it existed in the late 1960s in an earlier internal
Defense Department network called ARPAnet. There seemed to be little need in those
days to build an authenticated identity system to validate the sender of an email on a
closed system involving trusted parties. Since that time, there have been proposals put
forward to secure email by having the sender sign the mail in a trustworthy manner, but
those proposals achieved little market traction owing to lack of market demand,

24
Bank of England, “Financial Stability Report,” July 2015, Table A.10, p. 32, at
http://www.bankofengland.co.uk/publications/Documents/fsr/2015/fsrfull1507.pdf, accessed January
6, 2017.
25
Verizon, “2016 Data Breach Investigations Report,” p. 18, available at
http://www.verizonenterprise.com/verizon-insights-lab/dbir/2016/, accessed December 24, 2016.


engineering complexity, development costs, disagreements about the correct approach,
the lack of an institution that could exercise acceptable global leadership, and so on.
There would also be little if any market advantage to incurring these costs if others failed
to follow. These issues are not technical.
The vulnerabilities at all levels of the cyber environment have been well known for
years, yet many firms fail to take basic security precautions. And it is still the case that a
large majority of intrusions are discovered by law enforcement and other third parties
and not by the enterprise that owns the network.26
Even among owners and operators of
critical infrastructure, decisions to expose their operations to these vulnerabilities have
repeatedly been made with little or no regard for the risks thus imposed on the
enterprise, let alone the risks imposed across the entire economy. Enterprises that expose
their operations to the Internet must accept Internet services as they find them, replete
with vulnerabilities, and protect themselves accordingly. Insofar as those enterprises are
regulated, the cost of doing so should be reflected in the rates they are permitted to
charge.
In short, profound network insecurity has persisted for twenty-five years for many
reasons. A problem this enduring in so fundamental an area demands concerted
attention. It also calls for concentrating resources devoted to research and development
efforts (R&D) into technologies and policies to make attacks more difficult and expensive
to launch and less difficult and expensive to combat.
Coordinating Research Policy
There has been no shortage in recent years of federal pleas for research into
critical infrastructure cybersecurity, but they have tended to remain general and
hortatory. In 2009, for example, the Department of Homeland Security (DHS) published “A
Roadmap for Cybersecurity Research” that identified an important problem set but did
not develop a research agenda to deal with it. In 2011, the National Science and
Technology Council (NSTC) articulated the need for federal spending in basic cybersecurity
research but was content to describe challenge areas (e.g., mobile security, creation of
trusted spaces, etc.) rather than specific areas for research.
In 2013 a presidential policy directive emphasized that research was a critical
aspect of achieving critical infrastructure security and resilience27
but was not specific. In

26
Verizon, 2016 DBIR, p. 11, fig. 9.
27
Resilience is the ability to operate at an acceptable, if suboptimal, level of performance in the face of
attack or failure. For a thoughtful exploration of this concept, see Harriet Goldman, “Building Secure,
Resilient Architectures for Cyber Mission Assurance,” Case 10-3301, MITRE Corp., 2010, at
https://pdfs.semanticscholar.org/911a/9c301359a0bcbdc3e49b2f7a04cf7eef14b2.pdf, accessed
January 5, 2017.


June 2014, a subcommittee of the NSTC issued a cogent statement of federal
cybersecurity research objectives, but did not identify a path to get there. Last year DHS
brought additional attention to the challenge with its R&D plan for research in this area,
but the plan did not go beyond a general statement of objectives. Reports and directives
from high levels of government are inevitably general, but lack of follow-through and
inattention to detail are not inevitable. At the agency level, specific but uncoordinated
research projects are underway to tackle technical cybersecurity problems. For example,
at the Defense Advanced Research Project Agency (DARPA) a project on Organically
Assured and Survivable Information Systems (OASIS) focuses on increasing fault tolerance
in systems and networks. But these programs are not coordinated, and many of the
general problems described in high-level government documents remain insufficiently
addressed, if addressed at all.
Against this background, the nation must devote substantial coordinated
resources (1) to identify the most salient risks to critical infrastructure networks, and (2)
to describe specific cybersecurity objectives that could reduce those risks and that could
be broken into manageable research projects. This is what IPRI and CIS have sought to do.

The Workshop Plan

IPRI and CIS convened four sector-specific workshops to study the challenge of a
coordinated research and policy plan, and later a fifth workshop to distill what we learned
from the first four. It was clear from the start that “critical infrastructure” had become too
broad a rubric to guide our work. In the United States, the term means “systems and
assets, whether physical or virtual, so vital to the United States that the incapacity or
destruction of such systems and assets would have a debilitating impact on security,
national economic security, national public health or safety, or any combination of those
matters.”28

Sixteen sectors have now been designated “critical.” We therefore selected four
sectors we deemed most critical,29
and scheduled the following day-long workshops, all
held in Cambridge, Massachusetts at MIT:

● Electricity October 8, 2015
● Finance November 5, 2015
● Communications December 3, 2015
● Oil and Natural Gas (ONG) February 8, 2016
● Final Workshop May 2, 2016

28
42 U.S.C. § 5195c (e).
29
Time constraints precluded an additional workshop on the transportation sector.


Participants came from key industry firms in the United States, Canada, Japan, and
Europe; from pertinent government offices, from MIT, and from Carnegie-Mellon
University. Most of the MIT participants and several of the industry and government
participants attended all the workshops. We limited attendance to twenty people at each
workshop and did not ask for prepared presentations. Instead we asked participants to do
three things:

1. Describe their most severe challenges in terms of systemic issues;
2. Describe the characteristics of a more secure environment for IT and the OT linked
to it; and
3. Identify the technical, political, and economic impediments to achieving those
characteristics.

Each workshop took on a dynamic of its own. We asked questions but did not limit
the topics of discussion. Not surprisingly, some industry participants had difficulty framing
questions in strategic terms, while some academic participants had difficulty framing
theoretical questions that were relevant to the concerns of the industry participants. Yet
each workshop produced spontaneous, lively discussions that served to frame and
sharpen issues. Although we asked participants to address the three questions just stated,
the output of each workshop fell into a simpler dyad: a consensus list of the greatest risks
to the sector, and a consensus list of most important challenges for the sector. Except for
certain regulatory issues, every major challenge was discussed in every workshop. If a
challenge appears in the account of one workshop but not another, that is because it
received the most emphasis in that workshop.

To create a research agenda, we convened a fifth workshop of twenty participants
selected from the previous workshops and presented them with a distillation of ideas
from the previous sessions. We asked them to identify the most critical challenges across
all sectors and to turn those challenges into questions amenable to research. The
outcome of that workshop formed the basis of the IRPI-CIS statement of the seven high-
level challenges and the related recommendations and research questions in this report.


The Sector-Specific Workshops

1. Electricity Sector Workshop

Electricity sits at the base of any modern society’s operational structure. Nearly all
economic and social activity depends on it. Not surprisingly, the risk most feared in this
workshop, even more than loss of information, was disruption of service.

The electricity sector operates in a unique and complex regulatory environment
and displays striking internal differences, especially between the larger firms and the
smaller enterprises and cooperatives. Electricity transmission in the United States30
is
governed by federal law, but delivery is regulated by the fifty states and the territories in
inconsistent ways. As a general matter, regulated entities are entitled to a specified rate
of return on expenditures allowed into their rate base, as determined by their regulator.
They therefore have an incentive to make expenditures allowable into that base.
According to our industry participants, state regulation has historically been consistent in
its emphasis on rate regulation, which is a politically sensitive topic, and on safety.
Expenditures calculated to lower rates (such as software designed to create efficiencies)
or to improve safety are favored, they said. In contrast, network security has not been a
regulatory focus, and some participants asserted that capital expenditures necessary to
defend digital systems are more difficult candidates for regulatory approval. Because of
the asserted difficulty of assigning a return on investments in network security, such
expenditures were also more difficult candidates for corporate approval, according to
these participants. These statements should be verified because, if true, these factors,
together with the long lifespan of much of the sector’s OT, would impede the adoption of
needed security measures.

The Most Severe Risks

Risk 1: Risk from aging operating systems retrofitted with digital controls.

Most participants believed the most important risk factor for their sector was the
networking of aging valves, pumps, and other hardware that were designed to be
physically isolated and locked up, but which are now accessible remotely. Many of these
operating components were twenty or more years old. They now form parts of systems
that were retrofitted (“cobbled together”) to be electronically accessible through
acquisition programs that failed to take the resulting vulnerabilities into account. A
participant compared the state of the industry to the Office of Personnel Management,
which had digitized old systems without understanding the vulnerabilities thus created.

30
The U.S. electric grid is better described as being part of the North American electric grid. There are
many dependencies at the grid level between the U.S. and Canada.


Participants also stated that no one fully understood the extent to which the
electricity industry is tightly coupled with other sectors, and therefore did not sufficiently
understand the risk of catastrophic, macroeconomic failure. There was support for the
view that the Department of Energy should be more concerned about disruptions lasting
longer than two to three weeks.

Risk 2: Risk from third-party access.

One participant identified his company’s chief risk as unauthorized external access
to networks and systems owing to the extension of access privileges to third parties,
mostly vendors and other contractors. All agreed this was a significant risk factor. Some
doubted whether meaningful network perimeters still exist. In some cases, companies
required dual-factor identification and the use of a VPN to engage in remote
maintenance, but if the threat arose in a trusted vendor’s system, as some thought likely,
those steps did not help.

Data centers and the increasingly ubiquitous Internet of Things (“IoT”) also created
third-party risk. The IoT created an attack surface that was huge and expanding
dramatically, and many of the connected devices related to energy consumption and had
little or no security designed into them. If attacked, these devices could cause localized
failure and be used to steal customer information. They could also be organized into
botnets to attack any sector of the economy. That observation has since been borne
out.31

Risk 3: Risk Created by Regulatory Emphasis on Compliance versus Security.

Participants stated there was a confusion among many executives and regulators
about the difference between compliance with published standards and adequate
security. That confusion is not restricted to this sector. In contrast, no such confusion
exists among security professionals, who understand that compliance certifications are a
necessary condition of doing business but insufficient because they do not adequately
address constantly changing risks. Some participants also stated that the basic compliance
standard issued by the North American Electric Reliability Corporation, known as the
“NERC CIP,” compared unfavorably to standards issued by the Payment Card Industry.
Compliance is check-list oriented and gives a false impression of security. Participants also
emphasized cultural factors, noting that the oil-and-gas sector’s concerted emphasis on
physical safety may be a model for an emphasis on security.

31
David E. Sanger and Nicole Perlroth, “A New Era of Internet Attacks Powered by Everyday Devices ,”
New York Times, October 22, 2016, at http://www.nytimes.com/2016/10/23/us/politics/a-new-era-of-
internet-attacks-powered-by-everyday-devices.html?_r=0, accessed October 25, 2016.


The Challenges

The electricity workshop identified high-level security objectives supported, in
most cases, by more detailed objectives necessary to achieve them. Most of the identified
challenges were economic, commercial, and legal rather than technical. In nearly all cases,
however, meeting the objectives would require a substantial effort simply to gather the
data necessary for high-quality analytics. Obtaining the necessary data in ways that did
not create additional risk for the data provider would itself be a significant challenge.

Challenge 1: Quantifying risk at the enterprise, sectoral, and macroeconomic levels.

There was general agreement that quantifying risk was both difficult and
necessary. As one participant stated, a dollar spent on “vegetation management”
(trimming trees) was more valuable to his company’s board than a dollar spent on
cybersecurity, because its effect could be measured, whereas network risk could not.
Participants also stated that baselining risk – that is, describing the current state of a
network – was difficult but necessary to quantify risk. One participant stated that many
utilities do not even own their own data, which would be required for risk analysis,
intelligence gathering,32 and prediction.

Challenge 2: Measuring and reducing intra-sector and cross-sector fragilities through
simulation-based, cross-sector exercises.

These fragilities were insufficiently understood. There are about 3000 utilities in
the United States, but seven utility holding companies serve about 70% of U.S.
customers.33 The level of operating and security sophistication in the market was not
uniform. More attention should be paid to IT/OT inter-connection risk across this
disparate market and to coordinating defenses. There was general agreement that the
electricity sector lagged the financial sector in this regard, and that sectors were tightly
coupled. Participants did not believe the country could detect a series of rolling, low-level
events that could precipitate a crisis. Participants broke this challenge into three parts:

a. Compile the data required for quality simulations. Exercises between the electric
and the financial sectors could yield major security gains, participants believed.
Various exercises coordinated by the Treasury Department and the Financial

32
The U.S. Department of Energy (DoE) has spearheaded an effort called the Cybersecurity Risk
Information Sharing Program, or CRISP, to share classified as well as unclassified information in this
sector. See letter of Patricia Hoffman, Assistant Secretary, DoE Office of Electricity Delivery and Energy
Reliability to Tom Fanning and Fred Gorbet, August 5, 2014, at
http://www.nerc.com/pa/CI/Resources/Documents/Department%20of%20Energy%20Letter%20-
%20Cybersecurity%20Risk%20Information%20Sharing%20Program%20(CRISP).pdf, accessed January 6,
2017.
33
Information courtesy of the Edison Electric Institute.


Services Sector Coordinating Council were a good model.34
But simulations
require large quantities of good data, which firms have been reluctant to share.
Utilities measure success based on reliability, safety, low costs, and consumer
satisfaction. What data would induce companies to add network security to this
list?35

b. Secure the participation of state, local, and federal governments in cross-sector
simulation exercises. A series of disaster exercises called Gridex now exists, but it
is limited to public-sector stakeholders. In the next scheduled exercises, planned
for the autumn of 2017, “participation is open only to registered utilities and
others specifically invited by the utility (e.g., vendors, local law enforcement).”36

Additional exercises should broaden participation in the public and private
sectors.

c. In a collaboration between MIT and industry, develop realistic scenarios for
simulation exercises.

Challenge 3: Creating a model for a rational regulatory scheme that would align
investment and security requirements with risk.

Many participants stated that prevailing regulatory regimes create intense
pressure to adopt software technology without any pressure to secure it. The following
specific steps toward creating a better model were proposed:

a. Perform a comparative analysis of state regulation of electric utilities in
Massachusetts, Rhode Island, and New York. An industry participant with
experience in these jurisdictions stated that studying their differences would be
enlightening.

b. Compare data integrity measures in the electric and financial sectors. The
financial sector was said to be intensely concerned with data integrity and was
more advanced than this sector in securing it.

34
See, e.g., Sean Waterman, “Bank regulators briefed on Treasury-led cyber drill,” FedScoop, July 20,
2016, at http://fedscoop.com/us-treasury-cybersecurity-drill-july-2016, accessed November 8, 2016;
U.S. Department of Treasury, “Joint Statement from the U.S. Department of The Treasury and Her
Majesty’s Treasury,” November 12, 2015, at https://www.treasury.gov/press-center/press-
releases/Pages/jl0262.aspx, accessed November 8, 2016.
35
A participant noted that the automobile industry had created massive cyber vulnerabilities in
vehicles, but that the industry is fixing them now because the potential liabilities could be very large.
Regulated utilities were said not to face a comparable risk.
36
NERC, “GRIDEX IV Frequently Asked Questions,” p. 1, December 2, 2016, available at
http://www.nerc.com/pa/CI/CIPOutreach/Pages/GridEX.aspx, accessed January 5, 2017.


c. Study nuclear regulation as a potential model for the regulation of non-nuclear
electricity. A participant stated that the nuclear industry went from a prescriptive
to a performance-based regulatory regime, recognizing that technologies were
advancing more quickly than regulation could keep up.

d. Optimize legal, regulatory, and tax, policy for security investment to maximize
investment incentives and place costs where they can be reflected in the price
of the goods and services produced. Existing regulatory schemes and tax policy
did not do this, according to participants. There was broad but not unanimous
support for the view that liability should play a greater role in driving better
network security, and that now it plays almost none.

Challenge 4: Supporting a market for simpler, less vulnerable technology.

The widespread use of field-programmable gate arrays37
and multi-purpose
controls were cases in point. Both were cheaper to produce than special-purpose devices
and were highly capable – but were therefore more vulnerable. Creating a market for
limited-purpose devices was seen as more of a political and economic challenge than a
technical one. In this regard, some participants wanted to explore the use of analog
devices within, or alongside digital systems, especially at end points.

Challenge 5: Improving human expertise in network management.

a. Identify the skill sets uniquely required in this sector and expand the talent
pool. There are not enough qualified operating engineers and computer
scientists who understand the challenges unique to the electricity sector.

b. Investigate the “Rickover Model” for the training and selection of navy
personnel for the nuclear submarine service. When the U.S. Navy created a
nuclear submarine service, Admiral Hyman Rickover required applicants to
complete a rigorous training regimen for admission to the service. Could that
model be adapted for security professionals in this or other sectors?

37
“A field-programmable gate array (FPGA) is an integrated circuit designed to be configured by a
customer or a designer after manufacturing …. FPGAs contain an array of programmable logic blocks,
and a hierarchy of reconfigurable interconnects that allow the blocks to be "wired together", like many
logic gates that can be inter-wired in different configurations.” “Field-programmable Gate Arrays,”
Wikipedia, at https://en.wikipedia.org/wiki/Field-programmable_gate_array, accessed December 12,
2016.


Challenge 6: Integrating the Management of IT and OT.

Each utility is different in the way it integrates, or fails to integrate, the
management of operating technology (OT) and information technology (IT). Some do not
converge until the corporate level; others converge much lower down. No one believed a
single governance model would be useful, but the group did believe that IT and OT have
substantially converged – at any rate, they have converged sufficiently so that operating
systems can now be attacked through IT systems. Management structures should reflect
that fact.

a. Unify security functions. In the view of many participants, someone in the
enterprise should have a view of the full scope of security threat, from wherever
they came. The group did not agree on that person’s proper title and reporting
responsibilities, but did agree that he or she should report to an officer of the
company and possibly to the board.

b. Optimize OT/IT replacement cycles, which are out of synch. OT in this sector
has historically been on replacement cycles of 15-25 years. In contrast, IT
measures technology generations in 3-5 years. These cycles should be studied
and optimized.


2. Financial Sector Workshop

The finance workshop identified three risks that were unique or especially severe
in the sector:

1. Data integrity risk;
2. Systemic risk to the financial system that may not be apparent when
considering enterprises or the sector in isolation; and
3. Third-party risk arising from the inability to alter long-term contractual
arrangements with other market participants.

The financial sector also shares risks common to critical infrastructure, though it
has the most advanced network defenses of any sector.


Risk 1: Data Integrity.

Risk to the integrity of financial data topped the list of our participants’ concerns.
Our economy is based on a system of accounts recording who owes what to whom at any
moment. Those accounts are digitized, and so are back-up systems. An attack that
destroyed or corrupted the accounts of a major financial institution could wreak
devastating economic havoc unless those accounts could be quickly and reliably
reconstituted. The risk extends beyond banks to securities exchanges, brokerage firms,
investment companies, clearing organizations, and other financial enterprises.

A sophisticated network attack could lock-up this sector. A logic bomb, for
example, could randomly delete system files. According to one participant, that has
already occurred, and it took time to understand what had happened and to fix it. But
disruption is only one risk that could arise form from data loss or corruption. A subtle,
more limited operation that corrupted the pricing of selected securities, for example,
could be used to manipulate markets, create illegal profits and losses, and drive parties
out of business.

Participants agreed that a slowly rolling attack on an institution might create more
havoc than an attack that brought the institution to an immediate halt, for which the
larger institutions prepare. A “low and slow” corruption of accounts would be difficult to
spot, and unless it were stopped quickly, it would infect back-up systems, too. The longer
it lasted, the more backup accounts would also be infected. Research that addressed this
risk would be of great value.


Risk 2: Systemic Risk from Tight Coupling Within and Across Sectors.

Participants were concerned about the cross-sector risk created by the tight
coupling of finance, energy, and telecommunications, but they were also concerned about
risk from tight coupling within their sector. Several participants agreed that financial
enterprises assume that in this space all parties are managing their own risks and that
systemic risk is therefore also being managed through the sector, but they doubted this is
true. Notwithstanding the perception that the level of cooperation in this sector is high,
these participants believed it was insufficient and that more collective action on
information sharing would be required to better protect the sector from attack. The
nuclear power industry was cited as an example. In that sector there was widespread
understanding that an adverse incident that affected any of them would adversely affect
them all. The financial sector was said not to be at that point.

In particular, several participants complained of poor network security among
competing institutions (“shirking”). They gave two examples: (1) competitors that sought
market advantage by saving money on network security, and (2) community banks that
lacked the financial and other resources to make themselves reasonably secure. As to the
latter, participants noted that the share of assets controlled by community banks
continues to fall, so some questioned the significance of this risk. Others noted that
imposing further regulation on these banks would accelerate consolidation in the banking
sector. However, that risk was not equally troubling to everyone present. A participant
noted that shirking was merely one aspect of the more general problem of consistent
standards. As institutions other than banks and SEC-regulated businesses became larger
players, the problem of inconsistent regulation would present a growing problem. Several
participants stressed that one should pay close attention to the application of regulatory
standards as well as to their content when assessing consistency.

Risk 3: Contractual Risk from Long-Term Third-Party Contracts.

Long-term contracts with other institutions (which some participants called
“locked handshakes”) were a special example of risky intra-sector coupling. The example
given involved payment processors, which allegedly employ hard, pre-set passwords that
are not regularly rotated, if rotated at all. That kind of arrangement was said to lock in
network access rights of third-parties with allegedly poor security. These contracts were
said to allocate risk in ways that participants believed were unfair and that were not
foreseen when the contracts were made. These contracts can have terms of twenty years,
and many were made before the sector fully came to grips with network risk. These
assertions should be tested empirically. However, industry participants believed this risk
was real, that the sector needed a means to force the renegotiation of these contracts,
and that quantifying the problem would be helpful. We detected a willingness among
several industry participants to favor a regulatory solution to this issue, and one of them
specifically suggested that the issue could be of interest to the Federal Trade Commission


(which has recently used Section 5 of the FTC Act38
to address unfair as well as misleading
practices affecting network security). Another suggested that clearing agencies might be
able to provide leverage for achieving higher security levels. In evaluating these
contentions, attention must be paid to the competitive interests involved as well as to the
alleged security risks.

Risk 4: Difficulty of Identifying Malicious Actors.

The difficulty of attributing behavior to malicious actors is an aspect of the identity
management problem common to every sector, but our participants stressed the
challenge of ascertaining internal as opposed to external identities. And they were
concerned with controlling administrative privileges because most hacks they dealt with
involved abuse of administrator access. Some participants said that machines also have
identities and privileges, and that managing identities was easier for people than for
machines. Several participants stated, without dissent, that “operator risk” – that is,
insider threat from malicious or simply negligent behavior – was a medium, not a low,
probability. Some participants agreed that the government’s unsuccessful efforts
regarding trusted identities illustrated the difficulty of accomplishing anything
comprehensive in this space.

The Challenges

Challenge 1: Enhancing the integrity of backup systems.

A slowly evolving attack could be a bigger threat to financial institutions than an
attack aimed at a sudden network collapse because it would not be discovered as quickly
– and possibly not until backup systems had been infected. Participants were particularly
interested in the possible applicability of blockchain technology to their systems and the
status of blockchain research to the latency problem (that is, the time required to
complete a communication or transaction). Some participant firms are investing in
blockchain research.

38
15 U.S.C. §§ 41-58, as amended. The Commission is a consumer protection agency, not a financial
regulator. It considers three factors in determining whether a practice violates the prohibition on
unfair consumer practices: (1) whether the practice injures consumers; (2) whether it violates
established public policy; (3) whether it is unethical or unscrupulous.” FTC, “FTC Policy Statement on
Unfairness,” December 17, 1980, accessed November 16, 2016.


Challenge 2: Identifying and reducing cross-sector risk through joint cross-sector
exercises.

Robust joint exercises using sophisticated data would help illuminate the risk from
the tight coupling of power, finance, and telecommunications. These exercises would
elucidate intra-sector and cross-sector vulnerabilities and would benefit all participating
sectors. They would also highlight sectoral differences about the priorities given to
availability, integrity, confidentiality – another area for potential research.

Challenge 3: Improving identity management consistent with privacy concerns.

a. Among Communicants

The tension between privacy and identity management among communicants
concerned many participants, but there was widespread agreement that it is
important to focus on the specific information fields that would be most useful,
and then to determine whether and how that data can be shared consistent with
EU and US law. Several participants asserted that EU law made it more difficult
to identify both malware and malicious actors in their systems.
A non-industry participant stated that banks and credit card companies are not
using in their own networks the kinds of data-driven identity management/risk
flagging techniques they employ to monitor credit risk. It would be useful to
know whether, why, and to what extent this may be true.

b. Among Providers

It is technically simple to divert large amounts of traffic when it is “handed off”
from one service provider to another. This has occurred several times. These
hand-offs occur at border gateways, following border gateway protocols (BGP).
These protocols are weak, which is to say that identity assurance39
is weak at the
BGP level as well as at the level of individual communications. Traffic diversion
could cripple communications, and although it would be quickly discovered and
repaired, the delay in a crisis could be critical. A more secure version of BGP
exists, called BGPSEC, but few U.S. carriers have adopted it, presumably because
they do not expect a benefit from adoption that would offset its cost. What
economic or other factors impede the adoption of border gateway protocols
that would make it impossible, or substantially more difficult, to divert network
traffic? How can those factors be reduced or eliminated? Fixing this systemic
weakness would not appear to raise privacy concerns.

39
Machines, systems, and regions of the Internet, as well as persons, have identities.


Challenge 4: Containing the “Blast Radius” of Destructive Attacks.

It is now widely understood that malware cannot reliably be kept out of even very
sophisticated and well-run systems. The challenge was therefore to contain its effects – or
as one participant put it, to contain its “blast radius.” Participants returned several times
to this topic and were deeply interested in technical means of accomplishing this
objective (e.g., flexible segmentation and rapid reconstruction of networks).

Challenge 5: Modernizing the Regulatory Environment

Regulatory challenges fell into two groups: (i) creating flexible standards that
would improve security as well as guide compliance (a goal that may be as elusive in
theory as it has been in practice), and (ii) harmonizing regulations nationally and
internationally.

a. Flexible Standards

Industry participants stated that regulatory norms are not adapting to rapidly
changing technology and are rigid and costly without being effective. They noted
several instances where firms were compliant with applicable standards but were
penetrated anyway. They were interested in seeing flexible standards that would
evolve with technology and reduce risk when implemented – like a standard of
care. Participants referred to standards issued by the National Institute of
Standards and Technology (NIST) and the International Standards Organisation.40

These could evolve into enforceable standards of care, but legally binding
standards of care usually evolve through litigation; regulations are promulgated.

A non-industry participant stated that compliance and risk-based standards are
not necessarily in conflict, and that expecting government or a standards
organization to compel virtue was not realistic. He added that mandating red-
teaming forces threat-modeling. More broadly, he asked what success would look
like under a risk-based approach and suggested this could be a fruitful research
question. In this regard, participants would be interested to know whether
sectoral stress tests could be developed.

40
NIST, “Framework for Improving Critical Infrastructure Cybersecurity,” v. 1.0, February 12, 2014, at
021214.pdf, accessed January 7, 2017; ISO/IEC 27032: 2012 Information Technology – Security
Techniques – Guidelines for Cybersecurity, July 2012, at
http://www.iso27001security.com/html/27032.html, accessed January 7, 2017.


b. Harmonization

The Securities and Exchange Commission, the Commodities Futures Trading
Corporation, the Federal Financial Institutions Examination Council (among others)
regulate the financial sector in the U.S. Standards issued by the National Institute
of Standards and Technology (NIST) and the International Standards Organisation
also apply. These regulations and standards should be consistent and should
evolve rapidly. While there was broad agreement on this point, specific
inconsistencies were not discussed except to note that ISO 17799 was in effect for
about ten years before it was superseded.

We heard conflicting views from industry on the issue of global harmonization.
Some said that standards of the Federal Financial Institutions Examination Council
are already fairly global and that global convergence would have to occur; others
stated that the convergence of physical and logical security is making global
regulatory convergence ever more difficult. Regulatory differences between
Europe and the United States over the security of cloud computing remained a
problem. This was particularly troubling because as the number of sensors
expands dramatically through the IoT, the information they generate can be
managed only through cloud services based on big data analytics and machine
learning. Could research on the security of encryption in cloud computing be
useful in achieving international harmonization? Again, securing the data required
to do this research would be a challenge.

Challenge 6: Re-establishing trust in U.S. global leadership.

Since the Snowden disclosures the United States is no longer the most trusted
actor in network space, even among its allies. There is no trusted “sentry on the Roman
road.” Regaining international trust, especially among the Western democracies and
Japan, was a strategic challenge of high importance to the financial sector.

Challenge 7: Assessing Portfolio Risk from Information Insecurity.

This was a topic of widespread interest but not extended discussion. Whether it
could be accomplished without aggregating the risks created by weaknesses in the
systems of each company in a portfolio would be an interesting research question. Most
participants believed cyber risk was not factored into the pricing of securities.


Challenge 8: Identifying unauthorized executable code.

The architecture of the Internet – and the architecture of the hardware and
software that run on the Internet – permit executable code to reside among data files.
Indeed, embedding executable code within data files is a feature of some widely used
systems, making those systems harder to secure. Unauthorized or malicious code does
not identify itself as executable; it masquerades as a data file. In a typically large data file,
this kind of malicious insertion is therefore extremely difficult to find and eliminate. Can
this weakness be eliminated or made less severe?

Challenge 9: Enhancing workforce development.

Technical education and talent management were problems. Filling the pipeline
with highly educated, trained network engineers and computer scientists was the
challenge. Education and training should begin early in life.

Challenge 10: Creating effective cyber deterrence.

Is deterrence working? Against whom? Participants had differing views. Some
thought it was playing no role in bringing stability to networks because highly
sophisticated nation-state actors were operating in our networks. Others noted that
nation-states could wreak havoc with one another but had not done it, which suggested
that nation-states were indeed being deterred from escalating exploitation to
destruction attacks. However, there was broad agreement that fringe states such as
North Korea, Iran, and Syria might not be similarly deterred and that containing their
attacks would be a challenge going forward.

Challenge 11: Designing systems that do not fail silently.

A non-industry participant stated that one of our most important and difficult
challenges was to make “silent failure” impossible. He meant that when a system failed,
its operators should know immediately that it had failed. They should also be able to
determine why it had failed. To do that, we had to learn how to interrogate algorithms.
Absent the ability to do so, algorithms would become increasingly autonomous and
beyond human control. He suggested that critical sectors would probably differ on how
far autonomy should go, and that research on that point would be useful.


3. Communications Sector Workshop

This workshop was unique in focusing on the risks of managing networks that are
too complex to be understood and whose states cannot be known from one moment to
the next. Here again this risk not unique to this sector, but it was articulated only in this
workshop. The risk feared most in this sector, even more than data security, was
catastrophic failure. Many conditions created risk, but two conditions created the most
concern: (1) systems too complex to understand, and (2) loss of perimeter control.


Risk 1: Risk from indeterminate network state.

Over the years, communication hardware has grown massive and software even
more so. Once systems attained sufficient size – wherever that threshold might be, the
industry had crossed it – participants agreed that systems were inherently unstable and it
was difficult to know much about them. That was unnerving. Communications systems
were so complex that neither the firms that owned them nor their vendors fully
understood them. There was agreement, however, that software-defined networks
(SDNs) offered advantages in managing complex networks. Yet there were differences
expressed as to the nature of SDNs. One participant stated that they enabled
decentralized control and were therefore more resilient even if many nodes were taken
out. Another stated that SDNs pulled the network control algorithms out of the routers
and into a controller that was conceptually centralized. But it was agreed that, through
application programing interfaces, SDNs offered significant cost savings and could be
reconfigured or rebuilt swiftly even while under attack. No one doubted they would
become prevalent in this sector, but there was disagreement about whether SDNs were
reducing complexity.

Risk 2: Risk from third-party access and porous network perimeters.

In common with other sectors, industry participants saw widespread third-party
access with consequent loss of perimeter control as a major risk. The communication
sector is experiencing a major perimeter expansion through cloud computing, network
function virtualization, vendor access, and the IoT. The very notion of a perimeter was in
question. More physical devices on the network made the stakes much higher. Some
industry participants said they were engaged in constant electronic warfare in the
military sense.

Interoperability with legacy systems, both internal and external, was part of the
problem. Internally, there would always be some subsystems that were more secure


than others. Externally, some legacy systems were the equivalent of bad neighborhoods
that packets had to traverse.

The Challenges

Challenge 1: Communications networks should be harder to disable.

a. Failure in any part of the network should be evident. Failures were inevitable and
should be planned for. But no failure should be silent to the operator. One
participant referred to the “Chaos Monkey” approach to testing (in which various
elements of the system are turned off to see what happens), and said that the way
to know if something would fail silently is to make it fail. An industry participant
stated that in the public health arena, the reporting of certain diseases was
mandatory, and that this sector should adopt a similar policy.

b. Legacy systems should be retired on accelerated schedules. Tax incentives,
regulatory measures, and better internal risk-assessment could all be helpful in
achieving this goal.

c. Software defined networks (SDNs) should be simpler as well as prevalent.
Participants anticipated that network management and control would become
highly distributed as a mean of increasing resiliency and that systems would
increasingly manage themselves. They also posed three questions relating to
network management. First, they asked whether blockchain technology could help
manage international networks by keeping a record of all changes to a network,
including the changes that the network made to itself? Second, they asked
whether SDNs had the potential to improve the confidentiality of communications
– or make them worse. Third, they asked how SDNs could aid in segmenting
networks in real time to isolate the effects of a malicious intrusion.

d. Some critical facilities should be isolated from publicly accessible networks.
Isolation was a relative concept. Total isolation was not possible, as the Stuxnet
attacks on Iranian centrifuges demonstrated. With that understanding,
participants agreed that a small but undefined number of critical facilities should
not be “public facing” but should operate on virtual private networks (VPNs), with
air gaps and significant access barriers.

e. The technological monoculture should be more diverse, and its components
should be designed with security in mind. Participants saw technological diversity
as a desirable goal but did not envision a path to get there. They also noted that
diversity would require a standards-based architecture to support it. Some were
interested in the possible use of analog technology at some points to contain
system failure. However, all participants believed that the costs and benefits of


innovation should be assessed with security in mind, which is not now the case. It
was shocking that we still have injection attacks, for example. Many participants
believed that a more robust liability or regulatory regime would be required to
make vendors design for security, but there was no consensus on whether tort
liability would be welcomed, though it was probably more desirable than most
regulatory approaches.

Challenge 2: Regulations and standards should be evidence-based and flexible.

Industry participants severely criticized the current regulatory regime in their
sector. Compliance with current standards did not lead to better security and was said to
be “a joke” and “a race to the bottom.” There were frequent fines for low level
violations, and these fines were taken from security budgets. The net effect was a
reduction in security spending. Industry participants wished instead that regulators took
steps to ensure that security budgets were increased.

Several participants believed that regulators in the banking sector did a better job
of devising and enforcing reasonable standards. In any case, all industry participants
believed that (1) communications regulations should be dynamic and evidence-based, (2)
regulation should incentivize discovery of vulnerabilities and penetrations, rather than
the reverse, and (3) disclosure to the regulator should not be penalized (as in the
Aviation Safety Reporting System). There was also support for loosened rules in
emergencies based on prioritized threats. A participant suggested that universities could
play an important role in designing a better regulatory scheme by bringing economic and
legal expertise to bear on the challenge.

Like the electricity sector, the communications sector supports all the others.
Historically the various sectors have been regulated by different agencies that focused on
different issues. But now these agencies have begun to create cybersecurity regimes,
which are not consistent. Harmonization would be useful. One participant stated there
were simply too many regulatory and compliance standards and that the federal
government should recognize only the NIST standards.

Challenge 3: Continuity-of-operation planning should be widespread and robust for
critical infrastructure and critical resource sectors.

Participants said that in every national level tabletop exercise, the participants
assumed that communications would be available. This assumption was unrealistic and
should be abandoned. Participants wanted to see systematic use of sophisticated
modeling and simulations to anticipate and train for attacks. This prescription was
common to all workshops, and so was an acknowledgement of the impediment: i.e., the
lack of sufficient data to perform robust simulations and to create good models. Some


participants proposed that MIT play a role in generating models, but the need for data
remained.

Apropos of data sharing, Industry participants were clear that privacy concerns
must be carefully balanced against security concerns. They also stated that privacy
concerns impeded necessary information sharing – though most admitted that
competitive concerns were an even bigger impediment to sharing.

Challenge 4: The government should confront the need for communication priorities in
case of national emergency.

Industry participants said they were having difficulty engaging government about
prioritizing critical systems. If communications were crippled in a disaster, it would be
essential to determine what sectors and firms would get priority service, but government
was said to be reluctant to make that determination. In a national emergency,
prioritization (also known as tiering) was thought to be inevitable. It would make sense,
for example, to give priority to a nuclear power plant or hospital over personal calls, but
participants from telecommunications firms said they had no guidance on how to
prioritize calls and, as common carriers, were required to treat all calls alike.

This challenge is not unique to this sector. For example, in a national emergency,
power might have to be rationed, but it is unclear that anyone has the authority to
require it (as opposed to the authority to “coordinate” with distributors). But the issue
arose only in this workshop. Cross-sector simulation exercises could illuminate the
consequences of this lack of authority.

Challenge 5: The nation should develop a national deterrence strategy.

Deterrence involves both making targets harder to cripple and exacting a price
from an attacker. Here the emphasis was on the latter. Participants were troubled that
attackers faced little likelihood of paying a price for attacking U.S. targets, that the
country had no discernible strategy for punishing attackers, and that the lack of
consequences was emboldening adversaries. They noted the successful cyber theft of
massive amounts of intellectual property but were even more concerned with the
prospect of destructive attacks such as those on Saudi Aramco in 2012 and on Sony in
2014. They predicted that such attacks would increase in the next two years. Playing
defense, they said, is not a sufficient strategy.


Challenge 6: The domain name system should be strengthened.

The domain name system, commonly referred to as the DNS, correlates plain-
language names for computers, websites, etc. with a numerical IP address. Thus, typing
“www.mit.csail.edu” in a browser takes you to 128.30.2.155. The DNS is weak and
insecure, which makes spoofing identities easy. There is wide agreement that adopting a
secure version of the DNS known as DNSSEC would bring a significant improvement in
the security of the Internet. Enterprises should be incentivized to move to the more
secure system.

Challenge 7: The cadre of highly qualified network engineers and computer scientists
with security expertise should be greatly expanded.

The need in this regard was urgent but could not be met in the short term. The
need was felt as strongly in the regulatory agencies as in industry.


4. Oil-and-Natural-Gas Sector Workshop

Like all sectors, the oil-and-natural-gas (“ONG”) industry faced the full array of
cybersecurity challenges, but it was chiefly concerned with risk to availability of service.
This risk arose chiefly from three conditions:

1. Unduly complex, general purpose technology;
2. The inability to swiftly detect malware; and
3. The uncertain ability to swiftly isolate the impact of compromise.

The ONG sector resembles the electricity sector because of their common
reliance on industrial operating technology, and both share cross-cutting network risks.
At the generation/extraction level, however, this sector enjoys a higher level of threat-
information sharing among the majors, and it absorbs new technology more quickly.


Risk 1: Operational risk created by unduly complex, general purpose technology.

Industry participants singled out insecure, general purpose controls as a supply-
chain risk to their operations. The components available from vendors had far more
functionality than they needed or asked for, and with superfluous functionality came
vulnerabilities. These participants wanted lean components with no more functionality
than needed for a particular type of task, but such components were not available in the
market. Vendors found it far more profitable to sell generic devices to a wide market.

Risk 2: Operational risk from the inability to swiftly detect malware.

Disguised executable code is easy to insert into a network and exceedingly difficult
to find. The risk of malicious executable code is enhanced by supply chain risk but is
separate from it. Some participants stated that their inability to detect malware faster
was also caused by their inability (i) to visualize their entire network at once and (ii) to
know what hardware and software were running on their network.

Risk 3: Operational risk from the inability to swiftly contain the impact of compromise.

Participants assumed that all malware could not be kept out of their systems. They
focused on the risk from the inability to compartment the malware’s impact. As in the
financial sector workshop, the image of containing the “blast radius” of the malware was
appealing. The question was how to quickly seal off a compromised area of a network.


The Challenges

Virtually every challenge addressed in the previous workshops was addressed in
this workshop too, but the following challenges received the most attention:

Challenge 1: Creating a security environment on the model of this sector’s successful
campaign to improve its safety environment.

ONG firms have been successful in fostering safety consciousness across the
industry and thereby driving down the number and severity of physical accidents.
Participants did not believe the industry had made the same commitment to network
security. They noted that cybersecurity and physical security and safety had largely
converged. Operations were controlled by digital networks. Network intrusions could be
used, and indeed have been used, to sabotage operations and thus threaten health and
safety. In most companies, however, electronic networks and physical operations were
not managed holistically, and several participants stated that the engineering culture they
confronted did not understand network security. They saw this as both a management
problem and a problem of company culture – not a technological problem.

Challenge 2: Creating a government-industry partnership to foster a supply chain that
produces simpler, less vulnerable components, especially industrial operating controls.

General purpose components came with superfluous functionality, and every
functionality created potential vulnerabilities. But general purpose components were
cheap and profitable. One participant stated that vendors charge maintenance fees, so
they benefit from the insecurities they create because they get paid to fix them. This is a
commercial, not a technological, problem, and participants saw no solution to it unless
the government would support demand for special purpose components for critical
sectors by becoming a more demanding buyer. Several participants thought that the
departments of defense, energy, and homeland security could play that role. Participants
also discussed the potential use of analog devices at key points in their networks but
generally believed it would be impossible. Doing so, they said, would mean losing real-
time remote monitoring capability. The digital “toothpaste was out of the tube,” one
participant said. The question remained, however, whether analog devices could serve as
fail-safe mechanisms working in parallel with digital systems.41

41
For example, at a conference at MIT’s Sloan School of Management in the autumn of 2016, an
executive of a major U.S. energy company stated that his company used analog pressure gauges in
pipelines that would override a malfunctioning digital pressure system and shut down the line.


Challenge 3: Automatically identifying unauthorized executables.

The challenge of swiftly identifying malware resolved itself into the challenge of
automatically identifying unauthorized executable code. This was an aspect of
participants’ demand for adaptive systems and cyber capabilities at scale, which they
believed would be possible only through machine learning and artificial intelligence.
Capabilities at scale, machine learning, and artificial intelligence would in turn be available
only through cloud services, which must be secure.

Challenge 4: Automatically neutralizing or containing the effects of system failure.

Containing cascading failure would require adaptive systems. Immediate visibility
of failure was a prerequisite of containing its effects. Participants also believed that
following a failure, systems had to be able to explain what went wrong, even if they had
not previously confronted the same circumstances. To do these things, systems had to be
capable of machine learning. “Patching on the fly,” “dynamic segmentation,” and “self-
repair” were phrases often heard. These aspirational capabilities could be realized only
through big data analytics, which would likely be available only through secure cloud
services.

Participants also believed that containing cascading failure would require limiting
common mode attacks at scale. Systems were too homogeneous within and across
sectors. An attack on one system could therefore be repeated successfully against many
other systems. They therefore saw heterogeneity as a goal.

Challenge 5: Encouraging an enforceable standard of care.

Many participants, including several industry participants, favored a legal standard
of care for software and equipment and possibly for certain operational activities such as
patching. They wanted enforceable standards, much as building codes are enforceable.
They also referred to the function of the Underwriters Laboratory in raising standards for
electrical appliances. Manufacturers followed these codes because they could be legally
liable if they did not and because their insurance carriers required them to do so. At the
same time, no industry participant favored mandating statutory or regulatory standards.
There was some support, however, for peer reviews of the kind used in the nuclear
industry.

Several participants stated that standards of any kind required a standard
vocabulary. For example, some participants refer to OT as “everything south of the
firewall.” Others define OT as anything that produces a physical output. “Failure,”
“compromise,” and “security” also required standard definitions.


Challenge 6: Accelerating and automating patch management.

Participants identified three different challenges relating to patching: (i)
prioritizing patches, which in turn implied (ii) measuring the relative risk of unpatched
vulnerabilities; and (iii) accelerating the patching process without adding new operational
risk. Currently, patching sometimes takes up to four months, which is far too long.

Challenge 7: Assuring memory safety.

One participant stated that computer scientists spend too much time addressing
individual vulnerabilities and not enough time addressing classes of vulnerabilities.
Memory safety (specifically, eliminating buffer overflow) was a case in point. Another
participant stated that this is not basically a technological problem; we know how to fix
this. Why does this class of vulnerabilities persist?

Challenge 8: Developing a rational, risk-based model for investment and compliance.

Participants believed that quantifying risk would help rationalize compliance
regimes as well as investment decisions. The challenge is broader than simply quantifying
aggregate system risk, however, because rational investment involves more than a
determination of how much money to spend; it also requires a determination of how to
spend it. Several participants believed that insurance carriers could provide more useful
requirements than government-mandated standards. Others said that economics
departments should consider focusing on security economics as a field of study.

Several industry participants called for more robust threat intelligence. One noted
that Congress had resisted funding for prediction markets, which could be useful, and that
MIT could play a helpful role in creating or encouraging those markets.

Many of the firms represented in the workshop were not cutting cybersecurity
spending, even as other IT spending is decreasing with the low price of oil. The vendor
participants said they were seeing increased revenue from cyber products.

Challenge 9: Increasing support for simulation-based complexity modeling and
capability maturity.

There was mixed support in this workshop for more information sharing. An Oil
and Natural Gas Cybersecurity Network already exists, and some participants were
reluctant to expand this trusted network. One participant’s company was already a
member of twelve information-sharing networks; that was enough. However, there was
no dissent from a proposal for more sophisticated crisis simulations, which require
massive amounts of high-quality data from the participants. Several participants


suggested that MIT could play a useful role in co-sponsoring simulation exercises and
might be a trusted repository for the required data.

March 2017

________________________________________________________________________________________

Joel Brenner was the principal author of this report, with the support and assistance of
IPRI’s Daniel Weitzner, Dr. David C. Clark, Professor Hal Abelson, Dr. Shirley Hung, Dr.
Taylor Reynolds, Melanie Robinson and Adam Conner-Simons (CSAIL) and CIS Professor
Kenneth Oye, CIS Director Professor Richard J. Samuels, CIS Executive Director Dr. John
Tirman, Michelle Nhuch and Dan Pomeroy. The rapporteurs for the workshops were Reid
Pauly and Rachel Tecott.

THE INTERNET POLICY RESEARCH INITIATIVE:
In 2014, MIT established the Internet Policy Research Initiative (IPRI). The Initiative brings
together resources from many departments and centers within MIT. Its mission is to work
with policy makers and technologists to increase the trustworthiness and effectiveness of
interconnected digital systems. Its tools are engineering and public policy research,
education, and engagement. IPRI is headquartered in MIT’s The Computer Science and
Artificial Intelligence Laboratory, which is the largest research laboratory at MIT and one
of the world’s most important centers of information technology research.

MIT Center for International Studies
The Center for International Studies (CIS) supports international research and education
at MIT. It is the home of MIT’s Security Studies Program; the MIT International Science &
Technology Initiative, its pioneering global education program; the Program on Emerging
Technologies; and seminars and research on migration, South Asia politics, the Middle
East, cybersecurity, nuclear weapons, and East Asia. The Center has traditionally been
aligned with the social sciences while also working with MIT’s premier science and
engineering scholars. CIS produces research that creatively addresses global issues while
helping to educate the next generation of global citizens.

Message to White House to take Cyber Security seriously - from MIT

More Related Content

What's hot (7)

Viewers also liked (20)

Similar to Message to White House to take Cyber Security seriously - from MIT (20)

More from David Sweigert (20)

Recently uploaded (20)

Message to White House to take Cyber Security seriously - from MIT