Anthropic's Leaked Federal AI Safety Draft Sparks Backlash

A discussion draft of federal AI safety legislation drafted by Anthropic, which had been shared with a handful of congressional offices, leaked over the weekend and drew days of fierce criticism from the AI safety community, putting Anthropic's safety-first brand at risk.

Confirmed

• Anthropic's head of federal affairs AJ Cimino confirmed the draft exists, describing it as legislative language translating Anthropic's "Advanced AI Framework" proposal; Anthropic also released a new frontier AI policy framework.
• The core controversy is the intervention threshold: a model must plausibly cause over $1 billion in losses or 50+ deaths/serious injuries before the government could block deployment. The draft would require third-party safety assessments and disclosure obligations, with limited federal authority to stop dangerous model releases.
• Watchdog Americans for Responsible Innovation criticized the lack of hard safety standards; after Politico's reporting, Transformer News argued the draft is far weaker than the Obernolte/Trahan FRONTIER Act and could make blocking dangerous releases legally harder than today.
• Law professor Stephen L. Casper called the draft "quite bad," accusing it of diluting key provisions of existing bills in a textbook case of regulatory capture, and flagging Dario Amodei's recent statements as inconsistent.
• Grady Booch, citing @Rahll, said the framework treats 50 lives and $1 billion in losses as an acceptable cost.
• Researcher David Manheim was more measured, noting legislative lobbying inevitably lags model iteration.
• Brad Carson—whose Public First is backed by Anthropic and Amodei—also criticized it, arguing companies should not write their own safety standards, while crediting Anthropic's serious engagement.
• Nathan Calvin wrote a long post asking "What is Anthropic for?" expressing confusion at its legislative feedback.
• Amid the fallout, OpenAI policy chief Chris Lehane urged Congress to back the Obernolte/Trahan FRONTIER Act.

Unconfirmed

• A Reddit poster speculated the $1B/50-casualty figures were a deliberate negotiating anchor; this remains personal speculation.
• Researcher Ethan Perez relayed @ajcimino's view that the preemption language was imprecise; full context of the internal disagreement is unclear.

Why it matters

• The episode surfaces the structural question of whether frontier AI companies should draft federal safety legislation themselves: critics call it regulatory capture, and even Anthropic-aligned figures like Brad Carson oppose corporate self-standard-setting.
• The leak lands as Congress weighs the FRONTIER Act and other bills; Anthropic's draft is seen as potentially diluting tougher provisions, directly shaping US federal AI safety legislation.
• It also dents Anthropic's safety-first brand and hands rival OpenAI an opening to push its preferred legislation.

2026-10-18 ~ 2026-10-20 · 14 related posts

Full story(3 episodes)→

• Episode 1: Anthropic's secretly pushed AI safety bill draws fire from White House and safety researchers(2026-10-17, 29 posts)
• Episode 2: Anthropic's Leaked Federal AI Safety Draft Sparks Backlash(2026-10-18, 14 posts)
• Episode 3: Leaked Anthropic Federal AI Bill Draft Slammed as Weaker Than FRONTIER Act(2026-10-20, 2 posts)

• Grady Booch slams Anthropic for tolerating up to 50 deaths and $1B damages per model — Grady_Booch · 2026-10-18
• [source] Anthropic confirms draft federal AI safety bill translates its Advanced AI Framework into legislative language — edwinarbus · 2026-10-19
• Anthropic researcher flags imprecise preemption language in AI policy draft — EthanJPerez · 2026-10-19
• Anthropic privately pitching Congress a bill: government can block an SI model only above $1B damage or 50 deaths — AIFlow_ML · 2026-10-19
• Anthropic proposes blocking AI models only over $1B damage or 50+ casualties — fsharpman · 2026-10-19
• OpenAI's Chris Lehane pushes Congress to act as Anthropic leak roils AI safety world — ShakeelHashim · 2026-10-20
• [source] Leaked Anthropic draft AI bill panned as weaker than FRONTIER Act, 'safety-first' pitch rings hollow — ShakeelHashim · 2026-10-20
• [source] Law professor slams leaked Anthropic draft AI bill as textbook regulatory capture — StephenLCasper · 2026-10-20
• Anthropic's leaked federal AI bill draft sparks regulatory capture criticism — davidmanheim · 2026-10-20
• Anthropic: block AI models only if they threaten $1B damage or 50+ deaths — Ok-Conversation3213 · 2026-10-20
• Brad Carson of Anthropic-backed Public First criticizes Anthropic's AI framework bill — ShakeelHashim · 2026-10-20
• Anthropic's draft federal AI safety bill falls short of binding standards, watchdog ARI says — wfithian · 2026-10-20
• Nathan Calvin's long post asks one of the biggest questions: what is Anthropic for? — ShakeelHashim · 2026-10-20
• Leaked Anthropic-circulated federal AI bill criticized for lacking minimum safety standards — ShakeelHashim · 2026-10-20