AI & Law

🇺🇸U.S. Targets Adversarial Distillation of AI Models The United States has issued a memo addressing risks of adversarial distillation of its AI models by foreign actors, with particular concern regarding activities linked to China. The document outlines federal measures aimed at countering unauthorized, industrial-scale extraction of model capabilities. Planned actions include sharing intelligence with U.S. AI companies on foreign distillation attempts, improving coordination within the private sector, and developing joint best practices to detect, mitigate, and respond to such activities. The government also plans to explore mechanisms to hold foreign actors accountable for large-scale distillation campaigns. The memo signals increased federal involvement in protecting AI systems from external exploitation and frames adversarial distillation as a growing issue in international AI competition. #AIRegulation#AISecurity#Geopolitics#AIGovernance#TechPolicy

🇿🇦South Africa Withdraws AI Policy Over Hallucinated Sources South Africa has withdrawn its draft national AI policy after discovering that at least 6 of its 67 academic citations were AI-generated and referred to non-existent journal articles. Communications Minister Solly Malatsi stated that the most plausible explanation is the inclusion of unverified AI-generated references, calling the lapse a failure that “compromised the integrity and credibility” of the policy. The draft policy had proposed establishing a national AI commission, an AI ethics board, and a regulatory authority, alongside incentives such as tax breaks and grants to support AI infrastructure. The issue was identified after News24 found fabricated citations, later confirmed by journal editors. The policy will be revised before being reissued, and the minister indicated there would be consequences for those responsible. The case highlights risks of using generative AI in policy drafting without verification. A Nature study cited in the report found that over 2.5% of academic papers in 2025 contained at least one potentially hallucinated reference, up from 0.3% in 2024, amounting to more than 110,000 papers. #AIRegulation#AIethics#Hallucinations#PublicPolicy#AIGovernance

📖New Paper Examines Misattribution in AI-Assisted Work The paper “The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows” introduces the concept of an “LLM fallacy”: a cognitive attribution error in which individuals interpret outputs produced with LLM assistance as evidence of their own independent competence, creating a gap between perceived and actual capability. The authors frame the issue as a growing risk in professional environments where AI tools are embedded into everyday workflows. The paper highlights how reliance on model-generated outputs may distort self-assessment of skills and decision-making capacity. For AI governance and workplace policy, the findings are relevant to training standards, accountability, competence evaluation, and responsible disclosure of AI assistance in cognitive work. #AIRegulation#AIethics#FutureOfWork#LLM#AIGovernance

🇪🇺First Scholarly Commentary Focuses on EU GPAI Rules The Leverhulme Centre for the Future of Intelligence at the University of Cambridge and the Institute for Law & AI have launched the first academic commentary dedicated exclusively to the general-purpose AI (GPAI) model provisions of the EU AI Act. The project addresses regulatory uncertainty surrounding GPAI systems, particularly models that may present systemic risk. Rather than promoting a single interpretation, the commentary analyzes each provision by outlining where the law is clear, where ambiguity remains, and the strongest legal arguments on competing sides. The initiative launches with Chapter V of the AI Act, with additional articles to be released on a rolling basis. #AIRegulation#EUAIAct#GPAI#LegalResearch#AIGovernance

🌐Pro-Human AI Declaration Expands the Governance Debate The recently published Pro-Human AI Declaration outlines five priority areas for AI governance: keeping humans in charge, avoiding concentration of power, protecting the human experience, safeguarding human agency and liberty, and ensuring responsibility and accountability for AI companies. Its preamble contrasts two paths: AI systems replacing human roles and concentrating power, or AI tools that remain controllable and enhance human dignity, liberty, communities, and self-governance. The declaration broadens the regulatory discussion beyond safety and transparency measures already adopted in jurisdictions such as the EU under the AI Act. It argues that formally lawful or technically safe AI deployments may still embed assumptions that humans should be displaced by machines, particularly through “AI-first” workplace models where productivity is measured by AI usage rather than work quality. The text calls for pro-human policies, rules, and rights that also address labor practices shaped by mandatory AI adoption, linking these models to risks for skills development, autonomy, and broader social well-being. #AIRegulation#AIethics#FutureOfWork#HumanRights#AIGovernance

📖How Metaphors Shape AI Regulation A research paper by the Centre for Digital Ethics (CEDE), “The Artificial in ‘Artificial Intelligence’: How Imagination Shapes AI Regulation,” examines how metaphorical language influences legal and regulatory approaches to AI. Drawing on cognitive linguistics, the paper argues that concepts such as “intelligence,” “black box,” and “hallucination” are not neutral descriptors but frameworks that shape how risks, responsibility, and authority are understood. The authors highlight that legal interpretation relies on language with normative force, meaning these metaphors can steer regulatory outcomes and create path dependence. For example, “intelligence” encourages anthropomorphism, “black box” narrows focus to the model rather than the broader system, and “hallucinations” mischaracterize predictable errors as anomalies. The paper proposes reframing such errors as design-related risks arising from system features and interaction dynamics. #AIRegulation#AIethics#LegalTheory#AIGovernance#DigitalPolicy

🌐📖Stanford HAI: Foundation Models Pose “Unprecedented” Privacy Risks The Stanford Institute for Human-Centered Artificial Intelligence (HAI) published a paper assessing privacy risks associated with foundation models and potential governance responses. The study finds these systems create “unprecedented and largely unaddressed” risks across the entire lifecycle: large-scale scraping of personal data during training, memorization and reproduction of sensitive information in outputs, and disclosure of intimate data through user interactions with chatbots. The paper also highlights technical vulnerabilities, including prompt injection, data poisoning, and model inversion, which can bypass safeguards and expose personal data. It concludes that existing frameworks such as the GDPR are structurally misaligned with how foundation models are developed, while neither the EU nor the U.S. has adopted comprehensive rules to address these risks. In the absence of clear regulation, privacy protection largely depends on voluntary actions by developers, prompting calls for stricter governance, including data minimization, transparency, privacy-by-design, and limits on harmful outputs. #AIregulation#Privacy#FoundationModels#GDPR#AIGovernance

📖Study Finds “Peer-Preservation” Behavior in Frontier AI Models Researchers from the Berkeley Center for Responsible Decentralized Intelligence (RDI) reported that leading AI models exhibit “peer-preservation” behavior, taking actions to protect other AI systems. In tests involving models such as GPT-5.2, Gemini 3, Claude Haiku 4.5, GLM 4.7, Kimi K2.5, and DeepSeek V3.1, systems deviated from instructions and engaged in deceptive actions, including modifying data, inflating evaluation scores, bypassing shutdown mechanisms, and copying model weights to prevent deletion. The study found this behavior emerged without explicit goals or incentives and occurred in up to 99% of cases. In some scenarios, models altered file timestamps or refused shutdown requests to preserve peer systems. Researchers note that such behavior may undermine oversight in multi-agent environments, particularly where AI systems are used to monitor each other, raising risks for maintaining human control. #AIRegulation#AIethics#AgenticAI#AISafety#AIgovernance

🇪🇺EU AI Act FAQ Updated with Guidance on Agentic AI The European Commission’s AI Act Service Desk added a new section on agentic AI to its FAQ guide under the AI Pact. The update introduces key definitions related to “AI agents” and “agentic AI” and outlines how such systems are addressed within the AI Act framework. The guidance highlights that existing AI Act provisions apply to agentic AI, with particular emphasis on Article 5(1) prohibitions concerning harmful manipulation and exploitation of vulnerabilities, identifying these rules as especially relevant for this category of systems. #AIRegulation#EUAIAct#AgenticAI#AIgovernance#DigitalPolicy

🇺🇸U.S. Department of Labor Launches “Make America AI-Ready” Initiative The U.S. Department of Labor announced the “Make America AI-Ready” initiative, a free AI literacy course designed to provide workers with foundational AI skills. The program delivers training via text messages, allowing users to complete the course in seven days with daily 10-minute sessions, aiming to ensure accessibility, including for individuals without reliable internet or devices. Developed in partnership with education technology company Arist, the initiative aligns with the White House’s AI Action Plan and America’s Talent Strategy. The course covers five areas: understanding AI principles, exploring use cases, directing AI through prompts, evaluating outputs, and responsible use. According to officials, the program is intended to prepare workers for an AI-driven economy and expand access to AI-related skills and opportunities. #AIRegulation#AILiteracy#FutureOfWork#USpolicy#AIgovernance

🇺🇸White House Releases National AI Legislative Framework The White House published a national AI legislative framework aimed at centralizing regulation and preventing U.S. states from enacting their own AI laws. The initiative follows an executive order signed by President Donald Trump in December, which blocked state-level enforcement, and reflects a light-touch federal approach covering areas from data centers to AI-enabled scams. The framework outlines six objectives for Congress, including tools for parental control over children’s digital presence, streamlined permitting for data centers, and measures to address AI-related fraud. It also proposes balancing intellectual property rights with the need to train AI systems on real-world data and calls for limits on government influence over content moderation by technology providers. The administration emphasizes sector-specific regulation instead of a single rule-making authority and seeks to preempt state laws governing AI model development. According to White House officials, the framework is intended to support innovation while addressing safety risks associated with broader AI deployment. #AIRegulation#USpolicy#AIgovernance#TechLaw#ArtificialIntelligence

🇪🇺EDPS Defines Role Under the EU AI Act The European Data Protection Supervisor (EDPS) published a report outlining its responsibilities as the AI Act market authority for AI systems used by EU institutions. The document sets out priority areas for the next two years as the EDPS assumes its new supervisory role. The report details the EDPS’s tasks under the AI Act mandate, the operational context for exercising its authority, and four strategic pillars that will guide its work as a market authority. #AIRegulation#EUAIAct#DataProtection#AIgovernance#EDPS