Regulation, legislation, safety frameworks, and the global rules shaping artificial intelligence
Signal AI covers AI policy with the same skepticism we bring to every story. We distinguish between proposed and enacted law, track lobbying efforts, and explain what actually matters to founders, engineers, and investors — without the breathless framing or partisan spin that dominates much policy coverage.
After years of negotiation, the world's first comprehensive AI regulation is legally binding. We break down which obligations apply when, which models are in scope, and where the Act leaves room for interpretation — and litigation.
The European Union's AI Act moves from text to implementation. National regulators gain authority to issue fines, and high-risk system providers must complete conformity assessments.
The AI Act's risk-based classification system places general-purpose AI models with systemic risk under the strictest obligations. Member states have 12 months to designate national competent authorities. Early enforcement focus is expected on biometric identification and employment screening systems.
The US Federal Trade Commission signals broader use of Section 5 authority against misleading AI claims, filing its first cases against consumer-facing health and financial applications.
Chair Lina Khan's statement emphasized that existing consumer protection law applies fully to AI products. The cases target specific deceptive claims about diagnostic accuracy and investment returns, not the underlying models themselves.
G7 trade ministers agree on a non-binding code of conduct for organizations developing advanced AI systems, intended to inform future binding agreements.
The code covers safety testing, red-teaming, and incident reporting. It is voluntary and lacks enforcement mechanisms, but participants described it as a foundation for future treaty negotiations.
Bipartisan Senate working group releases a discussion draft that would require pre-market review for frontier models above a compute threshold.
The draft faces significant opposition from both industry groups and civil liberties organizations. Key unresolved questions include the threshold definition, the role of NIST versus a new agency, and preemption of state laws.
China's Cyberspace Administration tightens disclosure requirements for generative AI services, mandating clearer labeling of synthetic content.
The updated rules require watermarking and metadata tagging for AI-generated images, video, and text. Domestic providers must file updated algorithm descriptions by year-end.
The UK government increases funding for pre-deployment safety testing, but critics question whether voluntary engagement from labs is sufficient without statutory backing.
The report stops short of recommending new legislation, instead urging courts to apply existing fair-use doctrine case by case. Rights holders and AI labs both claim partial victory.
Version 2.0 adds specific guidance for generative AI and dual-use foundation models. Federal contractors will be required to attest compliance starting in 2027.
The non-binding framework emphasizes interoperability with EU and US standards while accommodating varying levels of regulatory capacity across member states.
The new memorandum focuses on federal procurement rules, requiring agencies to document AI use cases and conduct impact assessments before deployment.
AI policy moves through multiple channels — legislation, regulation, executive action, litigation, and international negotiation — often simultaneously. Understanding the process helps distinguish signal from noise.
Bills are introduced in national legislatures, often following years of committee study, stakeholder hearings, and lobbying. Most proposed AI legislation never reaches a vote.
Existing agencies — FTC, SEC, NIST, national data protection authorities — apply current law to AI systems through guidance, rulemaking, and enforcement actions.
Presidential orders, cabinet directives, and procurement rules shape federal AI use quickly, though they can be reversed by future administrations.
Treaties, bilateral agreements, and multilateral codes create voluntary frameworks that may later harden into binding obligations through domestic implementation.
The gap between a proposed AI bill and an enforceable rule is where most of the real work — and most of the lobbying — happens.
— Signal AI Policy Desk
No comprehensive federal AI law yet. Enforcement relies on FTC, SEC, and sectoral regulators. Executive orders shape federal procurement and safety testing.
All US coverage →The world's first horizontal AI regulation is in effect. National regulators are being designated. Fines up to 7% of global turnover apply.
All EU coverage →No single AI statute. The UK relies on existing regulators and voluntary safety testing through the AI Safety Institute.
All UK coverage →CAC and MIIT share oversight. Rules focus on content moderation, data security, and synthetic media labeling. Draft measures on foundation models under review.
All China coverage →Brazil's AI bill awaits presidential signature. India's advisory approach remains non-binding. Several African nations begin drafting data protection and AI rules.
All global coverage →