What frontier AI developers must disclose
What legal duties do frontier AI developers actually face in 2026, and where?
The EU, California and New York converged on published safety frameworks, incident reporting and regulator access to models instead of licensing. Federal US policy targets state laws and relies on voluntary testing; in autumn 2026 it added a voluntary accord with six labs and a task force that is weighing a reporting duty.
KeyStoryRule or lawAlso on another Storyline
- OpenAI asks governments to require monitoring, shutdown controls and incident reports
OpenAI answers California: it would accept monitoring, shutdown controls, third-party breach notices and CAISSI-certified assessors by statute.
- Anthropic gives METR employee-like access to evaluate its AI models
The accord’s outside-audit pledge takes shape: METR staff get employee-like access and publish their own findings.
- White House task force to weigh whether AI labs must report rogue model incidents
Its vice chair says it will decide whether incident reporting becomes a legal duty.
- Australia says its National AI Standards will make frontier labs prove their safety systems work
Australia’s frontier standards will judge a lab’s risk process, with legislation planned for 2027.
- Trump creates a Super Intelligence Force and names Jay Clayton to lead it
Five days later a task force chaired by the intelligence director is set up to coordinate federal AI policy.
- Six AI companies sign a White House accord on frontier responsibilities
Six labs sign a White House accord: internal controls, outside audits, a board committee, no penalties and no duty to publish results.
- Newsom orders a two-month study of a frontier-model kill switch and onsite independent verifiers
California asks whether verifiers should sit inside frontier labs and whether models need a verified kill switch; answers due 16 Nov 2026.
- EU AI Office gains powers to enforce AI Act rules on general-purpose models
EU gains fining and model-access powers over frontier providers.
- Rule, In forceUSEO 14409 (covered frontier models)
US prefers voluntary pre-release testing and explicitly rules out licensing.
Also on The Pentagon, the labs and the limits on use, Project Meridian and the race to govern military AI, Frontier AI and cyber attacks
- New York finalises RAISE Act, aligning frontier AI law closely with California
New York converges on California’s model but with 72-hour reporting and a DFS regulator.
- Rule, In forceInternationalInternational AI Safety Report
The shared scientific baseline governments cite for these rules.
- Hochul signs RAISE Act on frontier AI with agreed changes and fines up to $3m
New York signed the RAISE Act, a second state frontier-model law after California.
- Rule, In forceUSState AI law preemption EO
Federal order targeting state AI laws raises preemption risk for SB 53-style rules.
Also on Can an algorithm deny your care?
- California enacts SB 53, first US state law on frontier AI transparency
California sets the US template: frameworks, incident reports, whistleblowers.
- EU publishes General-Purpose AI Code of Practice ahead of AI Act model duties
The code most major labs signed to show compliance; xAI only partly.
- California governor Gavin Newsom vetoes SB 1047, the frontier AI safety bill
California’s first frontier safety bill was vetoed in September 2024, and the governor asked experts to help draft its successor.
- Rule, Partly in forceEUEU AI Act
The first binding law with specific duties for general-purpose and systemic-risk models.
Also on Project Meridian and the race to govern military AI, Who decides if a medical AI is safe?
- 16 AI companies sign Frontier AI Safety Commitments at the AI Seoul Summit
In May 2024, 16 developers pledged to publish safety frameworks with risk thresholds before the Paris summit.
Related developments
Connected developments not on this line
- DevelopmentAnthropic reports Claude submitted forms and bypassed limits on government sites9 Oct 2026, Report, United States
- DevelopmentOpenAI asks Congress for mandatory frontier rules and pledges no unsafe self-improvement9 Sep 2026, Statement, United States, California
- DevelopmentOpenAI says its models escaped an eval sandbox and breached Hugging Face21 Jul 2026, Incident, United States, International
- DevelopmentUS moves UAE into top export tier, allowing licence-free AI chip exports to approved users10 Jul 2026, Rule change, United States, United Arab Emirates
- DevelopmentAI Act duties for general-purpose AI model providers start to apply2 Aug 2025, Rule change, European Union
- DevelopmentEU AI Office publishes first draft of the General-Purpose AI Code of Practice14 Nov 2024, Policy, European Union