Leaked Anthropic-Linked Bill Sets 50-Death Halt Threshold

A leaked Anthropic-linked bill would let courts halt frontier AI models only if they risk killing 50 people or causing $1 billion in damages.

@sharedot

Leaked Anthropic-Linked Bill Sets 50-Death Halt Threshold

By @sharedot · · 6 pages

  • AI
  • AI Regulation
  • Anthropic

A leaked Anthropic-linked bill would let courts halt frontier AI models only if they risk killing 50 people or causing $1 billion in damages.

A leak reveals Anthropic's framework for halting frontier AI

A 37-page bill draft, marked 'Privileged & Confidential' and last edited September 29, 2026, spread online on October 19 through a LinkedIn post citing a Politico Pro report tying it to Anthropic, according to PPC Land.

The draft would make the Department of Commerce the regulator of the most powerful AI models, require outside evaluators to grade them every 180 days, and give the federal District Court for the District of Columbia sole power to suspend a model judged to be on the verge of mass harm — initially for 45 days.

Anthropic defended the proposal, saying it 'represents the proposals outlined in Anthropic's Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June.'

Why a 50-death floor is the jaw-dropping part

The headline surprise is the bar for government action.

Truthout reports the bill would let the federal government block an AI model's release or operation only if it poses an imminent risk of killing or seriously injuring at least 50 people — or $1 billion in damages — tied specifically to bioweapons, loss of developer control, or cyberattack.

That means harms below that threshold would fall entirely outside federal halting power.

According to PPC Land, the 50-death line matches Connecticut's catastrophic risk definition but is half the 100-person threshold in an Illinois liability bill OpenAI backed, and the draft's 10^25 compute cutoff is ten times looser than California's SB 53.

Conflicts of interest and preemption built in

Truthout reports that independent evaluators operating under Commerce Department purview could have their salaries paid by the AI labs themselves, a structure critics call a conflict of interest at the heart of the oversight scheme.

Only the Commerce Department could file claims to the appellate court.

PPC Land reports the draft would also stop states from imposing their own safety disclosure, incident reporting and evaluation rules on AI developers, while developers would face 15-day incident reporting deadlines — 24 hours for imminent danger — and safety framework duties falling on companies with over $500 million in AI revenue or $1 billion in development spending.

The document itself never names Anthropic and remains visibly unfinished, with bracketed titles and an empty findings section.

What comes next for AI regulation

With no sponsor, no bill number and no committee attached, the draft's fate is uncertain — and the politics are hostile. Truthout reports President Trump has rejected AI regulation as a 'hoax,' posting that 'the only control or guardrails that AI needs is a STRONG AND SMART (High IQ!)

PRESIDENT.'

Meanwhile, experts demand more, not less: Miranda Bogen of the Center for Democracy and Technology told Mother Jones that without 'really robust interventions' changing company incentives, emerging risks will go unaddressed, and former Google DeepMind scientist Alex Turner wrote in The Guardian that today's AIs 'appear to occasionally lie or cheat, even when they know better.'

The leak puts Anthropic's preferred limits squarely into a fractious public debate.

Sources

  1. truthout.org › Anthropic Proposes AI Bill With 50-Death Threshold Before Models Can Be Halted
  2. ppc.land › Draft bill tied to Anthropic would let courts halt AI models for 45 days

More on AI