---
title: "Anthropic circulates draft US AI safety bill that would let a court halt a model posing imminent catastrophic risk"
url: https://anawords.com/story/anthropic-draft-ai-safety-bill-50-deaths-1-billion-threshold
published: 2026-10-22T17:49:16+05:30
updated: 2026-10-22T17:49:16+05:30
section: Policy & Regulation
author: Mayank Sahu
checked: sources checked by AI, approved by editor
version: 1
---

# Anthropic circulates draft US AI safety bill that would let a court halt a model posing imminent catastrophic risk

The draft sets a threshold of more than 50 deaths or $1 billion in damage, relies on registered independent evaluators, and gives only the Commerce Department power to seek a halt.

**In short:** Anthropic has been sharing a draft federal AI safety bill with US lawmakers. It would let the Commerce Department ask a federal judge in Washington, DC, to halt a frontier model’s release if registered independent evaluators find a foreseeable, imminent risk of more than 50 deaths or over $1 billion in damage. Politico first reported the draft on 16 October 2026.

## What the draft proposes

Anthropic has been circulating a draft bill to US lawmakers, Politico first reported on 16 October; TechCrunch and Gadget Review followed.[^1][^2] An Anthropic spokesperson told Politico the draft reflects the company’s Advanced AI Framework, which it has promoted to lawmakers since June.[^1][^2] It has no legal force unless Congress passes it.[^2]

## The threshold

Intervention would require a “foreseeable, imminent risk” from a frontier model that could kill or seriously injure more than 50 people, or cause more than $1 billion in damage.[^1][^2] The risk must come from one of three scenarios: creating or releasing a biological weapon, an AI-enabled cyberattack, or a model evading its developer’s controls.[^1][^2] TechCrunch notes the thresholds resemble California’s SB 53.[^1]

## Who decides

- **Independent evaluators** would register with the Commerce Department and assess frontier labs and their models. Labs could pay them, as long as payment does not depend on the results.[^1][^2]
- **72 hours:** if an evaluator finds an imminent risk, it must notify Commerce within 72 hours.[^1][^2]
- **A court:** Commerce would ask a federal judge in the US District Court for the District of Columbia to halt the model’s release or use.[^1][^2]
- **No other agency** could halt a model on its own, under the reported language.[^1][^2]

Gadget Review adds that labs would need safety frameworks, transparency reports for new models, whistleblower protections and incident reporting to Commerce, and that some state AI laws would be preempted.[^2]

## Open questions

Both outlets question whether lab-paid evaluators can stay independent.[^1][^2] TechCrunch notes the scenarios are some distance from current concerns such as agent-driven hacking.[^1] Anthropic did not respond to TechCrunch’s request for comment.[^1]

## What it means

This is one company’s proposal, not a law. Its main idea is that a narrow, high bar plus a court process could make emergency action both possible and predictable. Supporters of that approach say clear rules help companies plan; critics may say the bar is too high or that preempting state laws removes useful protections. Both views will shape any final bill.

For readers outside the US, including in India, where a national consultation on AI rules is expected, the draft is a useful reference for how a government might handle rare, extreme risks while leaving everyday harms to other rules.

## What to watch

- Whether any member of Congress introduces a bill based on the draft.
- Responses from other AI labs, state lawmakers and civil society groups.

Related: [India’s AI regulation consultation paper](/story/india-ai-regulation-consultation-paper-vaishnaw) and [the FTC’s questions to OpenAI and Anthropic on AI safety](/story/ftc-civil-investigative-demands-openai-anthropic-ai-safety).

*Disclosure: Anawords uses Claude, made by Anthropic, as one of its research tools. Anthropic has no say in our coverage.*

## Fact ledger

- [Reported] Anthropic has circulated a draft federal AI safety bill; Politico first reported it on 16 October 2026. (source 1,2)
- [Reported] The threshold is a foreseeable, imminent risk of more than 50 deaths or $1 billion in damage. (source 1,2)
- [Reported] Evaluators registered with Commerce must report an imminent risk within 72 hours. (source 1,2)
- [Reported] Commerce would ask a federal judge in Washington, DC, to halt the model. (source 1,2)
- [Unconfirmed] Some state AI laws would be preempted. (source 2)
- [Not known] Whether any lawmaker will introduce the bill.

## Sources

1. [TechCrunch: Anthropic pitches AI safety bill that may block a model if 50+ people could be killed](https://techcrunch.com/2026/10/20/anthropic-pitches-ai-safety-bill-that-may-block-a-model-if-50-people-could-be-killed/)
2. [Gadget Review: Anthropic’s proposed AI safety bill only kicks in at 50 deaths or $1 billion in damages](https://www.gadgetreview.com/anthropics-proposed-ai-safety-bill-only-kicks-in-at-50-deaths-or-1-billion-in-damages)