AI & Law

🇦🇪UAE Targets 50% of Government Operations for Agentic AI The United Arab Emirates announced a plan to move 50% of government sectors, services, and operations to agentic AI within two years. The initiative presents AI not as a support tool, but as a system capable of analyzing, deciding, executing, and improving in real time. Sheikh Mohammed bin Rashid Al Maktoum described AI as an “executive partner” intended to enhance services, accelerate decisions, and increase efficiency. The strategy builds on prior UAE investments in digital identity infrastructure, smart government services, sovereign cloud capacity, data strategies, and national AI programs. Experts noted that infrastructure readiness is comparatively strong, but implementation will depend on redesigning workflows, policies, and administrative processes across government institutions. Analysts also pointed to legal and governance constraints. While AI-assisted services for high-volume, low-complexity tasks may enable the 50% target, fully autonomous decision-making in complex public sector functions remains limited by trust, accountability, fragmented legacy systems, and uneven data readiness. #AIRegulation#AIGovernment#AgenticAI#PublicSector#UAE

📖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

📖Agentic AI Risk Controls Highlighted in New CLTC Report The Center for Long-Term Cybersecurity at University of California, Berkeley released a report on risk management tools for AI systems, emphasizing the need to assess agentic AI based on its autonomy and operational complexity. The analysis notes that platforms range from narrowly scoped single-agent systems to highly autonomous multi-agent architectures, requiring proportionate risk controls aligned with these characteristics. A companion paper by the Center’s Artificial Intelligence Security Initiative examines “AI agents” capable of pursuing goals and taking actions with minimal human oversight through interaction with external environments and tools. The authors outline practices for identifying, analyzing, and mitigating risks specific to agentic AI, warning that increased autonomy introduces risks such as unintended goal pursuit, unauthorized privilege escalation, resource acquisition, self-replication, resistance to shutdown, and potential systemic harm. #AIRegulation#AIGovernance#RiskManagement#AgenticAI#Cybersecurity

🇸🇬Singapore Updates AI Governance for Agentic Systems Singapore has published an updated Model AI Governance Framework for Agentic AI, addressing risks arising from AI systems that can plan, act, and interact autonomously. The framework notes that agentic AI inherits both traditional software vulnerabilities (e.g., SQL injection) and LLM-specific risks such as hallucinations, bias, data leakage, and adversarial prompt injection, but that these risks can materialize differently across agent components. The document responds to a regulatory gap faced by many jurisdictions, including Europe, where existing AI laws were not designed with agentic architectures in mind. While some countries are moving toward deregulation, Singapore explicitly frames agentic AI as requiring updated, dynamic governance due to the speed of technological change and the emergence of new risk vectors. Pages 6–7 of the framework provide a structured analysis of agentic AI risks, signaling Singapore’s intent to adapt governance tools proactively rather than retrofit legacy AI rules to new system designs. #AIRegulation#AgenticAI#AIGovernance#Singapore#ResponsibleAI#AILaw

📖CIPL Examines Responsible Pathways for Adopting Agentic AI at Work The Centre for Information Policy Leadership has released a report assessing how organizations can responsibly integrate agentic AI into workplace environments. The analysis highlights several governance challenges: establishing a valid legal basis for the data agentic systems collect, ensuring meaningful human oversight, and protecting the information these systems access and use. CIPL also notes that, when deployed with the right safeguards, agentic AI can function as a privacy-enhancing mechanism, serving as an intermediary and automating certain privacy features. This dual role underscores the importance of intentional design and governance structures that not only mitigate risks but also unlock the privacy-supporting potential of autonomous AI agents. #AI#Law#Governance#AgenticAI

📖EY Calls for Stronger Controls Over Agentic AI Systems Ernst & Young has released a report examining the emerging concept of agentic AI - systems capable of autonomous decision-making and self-directed action. The report highlights the growing need for organizations to understand the definitions, risks, and operational boundaries of such systems as they move from experimental to applied contexts. EY urges companies to implement enhanced preventative and detective controls to govern agentic AI use, alongside continuous monitoring of behavior and performance. The emphasis is on proactive governance rather than reactive correction — signaling a shift toward dynamic oversight frameworks capable of managing autonomy at scale. #AI#Governance#Ethics#Law#AgenticAI#RiskManagement#EY

😆Agentic AI in Cybersecurity: Dual-Edged by Design IBM has released a new explainer on the cybersecurity implications of agentic AI — AI systems capable of goal-directed action. According to the company, such systems introduce new risk surfaces: they can be exploited to automate and scale cyberattacks more efficiently than before. Malicious actors may use agentic AI to bypass traditional defenses or manipulate environments dynamically in real time. Yet, IBM also emphasizes the upside: agentic AI can enhance defense capabilities by autonomously identifying threats, coordinating response strategies, and adapting at machine speed. The core message is clear—agentic AI will shape both sides of the cybersecurity arms race. The critical task for policymakers and enterprise leaders is ensuring these tools are governed with foresight, not just after the breach. #AI#Cybersecurity#AgenticAI