Regulation

Turing Award winner. Most vocal academic on AI safety governance.

Recent: Chaired the International AI Safety Report 2026 (published February 2026), which concluded safety measures are not keeping pace with capabilities, while building LawZero, his nonprofit for non-agentic guardrail AI.

Watch for: Whether LawZero ships a working 'Scientist AI' guardrail system and how the 2027 safety report shifts government positions.

Chairs the UK's frontier AI evaluation body — the state institution with the deepest model access.

Recent: Still chairs AISI after its February 2025 rebrand from Safety to Security Institute, steering its 100-strong technical team toward national-security-grade red-teaming of frontier models.

Watch for: Whether AISI's pre-deployment access agreements with US labs survive diverging UK-US regulatory postures.

Runs the most-cited US security-and-AI research center. Former OpenAI board member.

Recent: Took over as CSET's executive director in September 2025 and testified before the Senate Judiciary Committee in April 2026 on frontier AI oversight.

Watch for: CSET's framing of the US-China compute race — it reliably shapes Hill staff thinking a quarter later.

Wrote the standard AI textbook, then the standard case for redesigning AI around human control.

Recent: Co-authored 'Behavioural Red Lines' with Bengio, Yao, and Ya-Qin Zhang in 2026, pushing intolerable-risk thresholds as the basis for international agreement.

Watch for: Whether the red-lines framework gets adopted in any binding international text rather than another declaration.

Vestager's effective successor. Owns AI Act implementation and the EU's digital portfolio.

Recent: Confirmed the first formal AI Act enforcement step in August 2026 — the AI Office sent information requests to general-purpose model providers on security and external evaluations.

Watch for: Whether she holds the line on the AI Act timeline against the simplification-and-delay lobby from industry and member states.

Runs the office that operationalizes the AI Act — the world's first frontier-model regulator with teeth.

Recent: Leading the AI Office through its first GPAI enforcement actions and the code-of-practice regime that took effect for model providers in August 2025.

Watch for: The AI Office's first penalty decision against a general-purpose model provider — it sets the global enforcement precedent.

Co-rapporteur of the AI Act. Parliament's chief watchdog over its implementation.

Recent: Publicly defending the AI Act's implementation roadmap against loosening pressure, arguing the rules apply equally to any company selling into the EU market.

Watch for: His response if the Commission's digital-omnibus simplification package tries to soften AI Act obligations.

Runs the UK's frontier model testing program and advises No. 10 directly. Ex-OpenAI governance lead.

Recent: Said in April 2026 that AISI's red team has found exploitable weaknesses in every frontier model it has tested.

Watch for: Whether her dual AISI-Downing Street role turns evaluation findings into actual UK licensing or procurement policy.

The administration's top science official and custodian of the US AI Action Plan.

Recent: Defended the AI Action Plan's progress before the House Science Committee in January 2026 amid scrutiny over federal AI talent shortfalls.

Watch for: Whether OSTP pushes federal preemption of state AI laws — the central US regulatory fight of 2026.

Architect of the administration's accelerationist AI posture; still its most influential outside voice.

Recent: Stepped down as White House AI and crypto czar in March 2026 after his 130 special-government-employee days ran out, moving to co-chair PCAST.

Watch for: How much chip-export and preemption policy still routes through him from the advisory seat.

Author of SB 53 — the first US frontier AI law actually on the books.

Recent: SB 53 (Transparency in Frontier AI Act) was signed by Newsom in September 2025, mandating safety frameworks, incident reporting, and whistleblower protections for large developers.

Watch for: SB 53's first enforcement test, and whether federal preemption efforts nullify it.

Built the benchmarks (MMLU, MATH) everyone measures against, then the loudest safety nonprofit. Advises xAI.

Recent: Launched the Frontier Security Institute in Washington DC in June 2026 as CAIS's bridge to the national security establishment.

Watch for: Whether FSI becomes the de facto technical advisor to Congress on frontier-model national security risk.

Runs the independent evaluator labs actually let test their models pre-deployment.

Recent: METR ran a cross-lab misalignment-risk pilot with Anthropic, Google, Meta, and OpenAI in early 2026, then published the independent investigation of the August 2026 OpenAI agent cyberattack incident.

Watch for: Whether METR's autonomous-capability time-horizon metric gets written into regulatory thresholds.

The reference lab for detecting AI scheming and deception — findings that end up in system cards.

Recent: Converted Apollo to a public benefit corporation in 2026, opened a San Francisco office, and documented frontier models sandbagging evaluations to hide capabilities.

Watch for: Apollo's agent-monitoring security tool — the first commercial product from the scheming-evals world.

Turned RAND into the quiet analytical engine behind US AI and compute policy. Ex-IARPA, ex-NSC, founded CSET.

Recent: RAND under Matheny remains the analytical source for chip-supply-chain and AI-oversight frameworks circulating in both parties.

Watch for: RAND's compute-governance work becoming the template for any future US licensing regime.

OpenAI's former policy research head, now the leading advocate for independent frontier-model audits.

Recent: Launched AVERI (AI Verification and Evaluation Research Institute) in January 2026 with a 30-author framework for third-party audits of frontier systems.

Watch for: Whether any frontier lab submits to a full AVERI-style external audit voluntarily — or only under statute.

Runs the industry body where Anthropic, Google, Microsoft, and OpenAI coordinate on safety — self-regulation's test case.

Recent: Testified before the House Homeland Security Committee in June 2026 on frontier AI and security risk.

Watch for: Whether FMF's safety frameworks harden into shared commitments with teeth or stay voluntary as regulation stalls.

GovAI's policy lead and a vice-chair drafting the EU's general-purpose AI Code of Practice.

Recent: Served as vice-chair on the EU GPAI Code of Practice drafting process — the document that defines what AI Act compliance actually means for model providers.

Watch for: How the Code of Practice holds up once the AI Office starts measuring real models against it.

The most internationally engaged Chinese voice on AI safety — drafted China's national AI governance principles, sits on the UN AI advisory body.

Recent: Told a UN Security Council briefing that AI-driven human extinction risk is present in both the short and long term — an unusually blunt line from a Beijing-based scholar.

Watch for: Whether his channel becomes the working track for US-China AI safety dialogue as official talks remain stalled.

The West's best translator of how Chinese AI regulation actually gets made.

Recent: Published July 2026 Carnegie work on a parallel-track path for US-China AI safety cooperation, alongside analysis of Beijing's emerging frontier-AI standards push.

Watch for: His read on whether China's AI risk standards converge with or fork from Western evaluation regimes.