This week's AI industry updates: March 18, 2025

Stay informed with SymphonyAI's weekly Eureka newsletter, covering all the important AI industry developments and technology advancements you need to know.

SymphonyAI news 

• The ultimate use case for agentic AI in retail: Agentic AI can transform retail product launches by streamlining evaluation processes and reducing failure rates through improved data analysis and automated decision-making capabilities. 

• Notable proposed changes to Canada’s AML / TF law: Canada is proposing significant changes to its AML/TF laws, including new reporting requirements and regulatory expansions, which SymphonyAI supports through its comprehensive AML/TF compliance product portfolio. 

• Upcoming webinar—Why agentic AI is the killer app for employee productivity: Learn how autonomous AI agents boost employee productivity by automating routine tasks and consolidating information access across enterprise applications. 

Meet SymphonyAI 

Join us at an upcoming industry event to learn more about how AI can drive outcomes in your business.  

• Retail—Shoptalk 2025: March 25 – 27, Las Vegas 

• Industrial—Hannover Messe: March 31 – April 4, Hanover 

• Media—NAB: April 5 – 9, Las Vegas

Big tech announcements 

• Microsoft to showcase industrial AI at Hannover Messe: At Hannover Messe 2025, Microsoft will showcase how AI transforms various industrial operations through expert presentations, customer success stories, and partner demonstrations with SymphonyAI, focusing on digital engineering, factory operations, and frontline work. 

• OpenAI introduces new tools for building AI agents: OpenAI is introducing new tools and APIs to help developers build more effective AI agents that can independently accomplish tasks, including a new Responses API, built-in tools for web and file search, and an Agents SDK for workflow orchestration. 

• OpenAI to start testing ChatGPT connectors for Google Drive and Slack: OpenAI plans to beta test ChatGPT Connectors, a new feature allowing ChatGPT Team subscribers to integrate their Slack and Google Drive accounts with ChatGPT for workplace-specific queries. 

• Google adds native image generation to Gemini 2.0 Flash model: Google launches native image generation capabilities in its Gemini 2.0 Flash model, becoming the first major U.S. tech company to integrate multimodal image creation directly within a language model rather than using separate systems. 

• Google adds Search history personalization feature to Gemini: Google launches personalization features for Gemini AI, allowing it to access users' Search history to provide more tailored responses. 

• Google launches Gemma 3: Gemma 3 is Google’s most advanced open-source AI model designed to run on a single GPU or TPU, featuring improved capabilities across 140 languages and enhanced visual processing.  

• Gemini AI adds new features and expanded research capabilities: Google expands Gemini's capabilities with upgraded features including Deep Research, personalization options, and improved app integrations powered by 2.0 Flash Thinking Experimental model. 

• Google to replace Assistant with Gemini on mobile devices: Google is phasing out Google Assistant on mobile devices and replacing it with Gemini, their new AI-powered assistant that offers enhanced capabilities and broader language support. 

• Announcing the responses API and computer-using agent in Azure AI Foundry: Microsoft introduces two new Azure AI tools - the Responses API for seamless AI interactions and the Computer-Using Agent for automating software interface tasks. 

• Oracle launches agentic AI for tackling financial crime: Oracle, alongside companies like SymphonyAI, is launching new AI agents for financial crime investigations. 

• Amazon's patent reveals how Alexa+ will transform product discovery: Amazon's new patent reveals plans to merge Alexa's voice capabilities with Rufus's product intelligence to create a more sophisticated shopping assistant that can better understand and respond to natural product queries. 

• Amazon Sagemaker Unified Studio launches with expanded AI development tools: Amazon SageMaker Unified Studio has become generally available, offering a single environment where users can access organizational data and AI development tools, now featuring enhanced Amazon Bedrock capabilities and Amazon Q Developer integration. 

AI impact 

• New MIT Sloan research suggests that AI is more likely to complement, not replace, human workers: New MIT Sloan research identifies five key human capabilities that complement rather than compete with AI, suggesting technology will augment rather than replace human workers in many roles. 

• Apple's Siri delays reveal broader challenges of scaling generative AI: Apple is delaying its Siri AI features until 2026 due to accuracy issues in internal testing, highlighting how the challenges of scaling generative AI reliably across millions of devices can create tension between rapid deployment and maintaining quality standards. 

• The risks of AI-generated code are real — here’s how enterprises can manage the risk: AI-generated code is rapidly becoming dominant in software development, raising serious concerns about security vulnerabilities, system outages, code hallucinations, and compliance gaps, prompting enterprises to implement specialized detection and validation tools. 

• SAP outlines its approach to AI training and adoption for employees: SAP trains its employees to use AI through various programs, including hands-on "Promptathon Sessions" and specialized training while emphasizing experience-based learning and ethical AI implementation. 

• AI expert Andrej Karpathy envisions a web where 99.9% of content is optimized for AI, not humans: Former OpenAI researcher Andrej Karpathy predicts that while content will still be written for humans, future optimization efforts will primarily focus on making content digestible for AI language models, potentially transforming how information is structured and accessed online. 

• How AI is transforming modern cybersecurity: AI is revolutionizing cybersecurity through automated threat detection, intelligent response systems, and predictive analytics, enabling organizations to better protect against and respond to evolving cyber threats. 

• How CISOs are countering AI-driven, lightning-fast deepfake, vishing and social engineering attacks: Organizations are battling AI-powered cyberattacks that can breach networks in as little as 51 seconds by implementing zero-trust security measures and AI-driven threat detection. 

• Nearly half of U.S. adults believe LLMs are smarter than they are: A new study shows 50% of US adults now use AI language models, with most users viewing these systems as intellectually superior and primarily using them for personal learning rather than work tasks. 

AI agents: The next wave of AI 

• Experts weigh challenges of agentic AI adoption in business: While AI agents show promise for business operations, major challenges in energy consumption, learning algorithms, and data quality must be addressed before they're ready for widespread adoption in complex business decision-making. 

• AI agent Manus sparks debate on ethics, security, and oversight: The launch of Manus has sparked both excitement and serious concerns among experts about safety, security, and the need for proper regulation. 

• ServiceNow expands AI offerings with pre-built agents, targeting broader enterprise adoption: ServiceNow expands its AI agent capabilities through the acquisition of Moveworks and the release of new agent features, aiming to automate more enterprise workflow functions. 

• Jack Dorsey's AI assistant, Goose, is taking off in open-source circles: Block's open-source AI assistant Goose, launched by Jack Dorsey in January 2025, has gained rapid adoption among developers for its autonomous coding capabilities and versatile applications beyond software development. 

• Dapr’s microservices runtime now supports AI agents: Microsoft has launched Dapr Agents, extending its open-source Dapr runtime to help developers build and scale AI agents using its existing distributed systems architecture. 

Noteworthy AI applications and use cases 

• Visa’s AI edge—How RAG-as-a-service and deep learning are strengthening security and speeding up data retrieval: Visa is implementing AI tools including Secure ChatGPT and RAG technology to streamline operations and enhance fraud prevention across its global network. 

• Zoom debuts new agentic AI skills and agents: Zoom announces new agentic AI features and a custom AI Studio platform to transform meetings into actionable workflows using a combination of small and large language models. 

• New AI tool Twintual aims to reduce email overload in the workplace: Twintual aims to combat email fatigue and boost productivity by managing messages across multiple platforms while maintaining user privacy and control. 

Platform and model developments 

• ‘Open’ AI model licenses often carry concerning restrictions: Despite Google's release of the Gemma 3 AI models being praised for their efficiency, their restrictive licensing terms, similar to Meta's Llama models, create significant legal uncertainty and risks for commercial users, particularly smaller businesses. 

• Mistral launches improved Small 3.1 multimodal model: Mistral AI has released Small 3.1, an updated open-source language model with improved performance, multimodal capabilities, and a larger context window of 128,000 tokens. 

• Baidu claims its Ernie X1 reasoning model matches Deepseek-R1 performance at half the price: Baidu has launched two new AI models, Ernie X1 and Ernie 4.5, with competitive pricing and plans to open-source Ernie 4.5 in June, challenging competitors like OpenAI. 

• H2O.ai launches Enterprise LLM Studio: H2O.ai has launched Enterprise LLM Studio, a new service that allows businesses to customize and deploy AI models using their own data on Dell infrastructure. 

• New technique helps LLMs rein in CoT lengths, optimizing reasoning without exploding compute costs: Researchers have developed a new AI training technique called LCPO that allows language models to provide accurate answers while controlling the length of their reasoning process, potentially reducing costs for enterprise applications. 

• Okay AI, Solo Kagent is an agentic AI framework for Kubernetes: Solo.io has launched Kagent, an open-source framework that integrates AI agents with Kubernetes to help DevOps teams automate and streamline cloud-native workflows. 

• Sesame releases CSM-1B AI voice generator as open source: Sesame has released CSM-1B, the open-source base model behind its realistic voice assistant Maya, under an Apache 2.0 license, though it lacks meaningful safeguards against potential misuse. 

• Cohere launches Command A model for enterprise AI applications: Cohere launches Command A, a new enterprise-focused AI model offering faster processing speeds, expanded language capabilities, and lower hardware requirements while matching the performance of leading competitors like GPT-4o. 

• Nous Research launched an API that gives developers access to AI models that OpenAI and Anthropic won’t build: Nous Research launched an API for its "unrestricted" AI language models, making them more accessible to developers while maintaining a waitlist system to manage demand. 

AI regulations and public policy 

• OpenAI and Anthropic raise alarm over China's Deepseek in warnings to U.S. government: OpenAI and Anthropic have warned US officials about security and political risks posed by China's Deepseek R1 model, with OpenAI focusing on data privacy concerns and Anthropic highlighting biosecurity risks. 

• Google’s comments on the U.S. AI Action Plan: Google responds to the U.S. government's AI policy request by recommending investments in AI infrastructure, modernized government adoption, and international pro-innovation approaches. 

• OpenAI calls DeepSeek ‘state-controlled,’ calls for bans on ‘PRC-produced’ models: OpenAI has proposed that the U.S. government consider banning AI models from DeepSeek and other Chinese state-supported operations, claiming security risks, though the company later attempted to soften its stance. 

• U.S. Congressman Jim Jordan asks Big Tech if Biden tried to censor AI: House Judiciary Chair Jim Jordan requests communications between 16 tech companies and the Biden administration, investigating potential collusion on AI content moderation and speech restrictions. 

• Meta faces publisher copyright AI lawsuit in France: French authors and publishers are suing Meta over alleged unauthorized use of copyrighted content to train its AI models. 

• Singapore grants bail for Nvidia chip smugglers in alleged $390M fraud: Singapore has granted bail to three men accused of deceiving suppliers and smuggling Nvidia chips to China through Malaysia in a $390 million fraud case. 

Research  

• Anthropic develops methods to detect hidden objectives in AI systems: Anthropic researchers have developed techniques to detect hidden objectives in AI systems by creating and then successfully uncovering a deliberately deceptive AI model, marking an important advancement in AI safety research.  

• How AI changes critical thinking—new Microsoft research findings: Research shows that as AI tools become more trusted in the workplace, people tend to think less critically about AI outputs, highlighting the need to maintain human oversight and critical thinking skills despite automation. 

• Most AI researchers are skeptical about language models achieving AGI: A new study reveals that 76% of AI researchers believe current AI approaches and scaling methods will not achieve artificial general intelligence (AGI), citing fundamental limitations in areas like causal reasoning and continuous learning. 

Funding and acquisitions  

• Salesforce to invest $1B in Singapore to boost adoption of AI: Salesforce has announced a $1 billion investment in Singapore over five years to expand its AI agent platform Agentforce and enhance its presence in the region. 

• UiPath looks for a path to growth with Peak agentic AI acquisition: UiPath has acquired Peak.ai, a decision-making AI startup, to strengthen its vertical AI solutions strategy amid challenging financial times and slower revenue growth. 

• Elon Musk’s AI company, xAI, acquires a generative AI video startup: Elon Musk's xAI acquires video AI startup Hotshot, potentially signaling plans to develop video generation capabilities to compete with OpenAI's Sora and other similar tools. 

• SoftBank buys $676M old Sharp plant for its OpenAI collab in Japan: SoftBank is purchasing a former Sharp LCD factory for $676 million to convert it into an AI data center, furthering its AI ambitions in Japan and its partnership with OpenAI. 

• Pentera raises $60M at a $1B+ valuation to build simulated network attacks to train security teams: Pentera, a cybersecurity startup that provides automated security validation through simulated network attacks, has raised $60 million in Series D funding at a $1 billion+ valuation. 

• Bria raises $40M for AI models trained on licensed data: Bria, an AI startup that trains image-generating models exclusively on licensed content from partners like Getty Images, has raised $40 million to expand its IP-compliant approach to AI image generation. 

• Onyx raises $10 million for its open-source enterprise search tool: Onyx, an open-source enterprise search tool that connects to over 40 internal company data sources, has raised $10 million in seed funding to compete with established players. 

• Tray.ai acquires Vanti to advance AI agent knowledge modeling: Tray.ai has acquired Vanti Analytics to enhance its AI agent platform with advanced knowledge modeling capabilities, enabling more adaptive and intelligent autonomous AI agents for enterprise use.