Anthropic is circulating a draft AI bill among US lawmakers that would require frontier developers to test their most powerful models and could let the federal government block a model’s release or use in narrowly defined danger scenarios. It would also override some state AI laws, making this a consequential proposal, not merely another call for “responsible innovation”.
Anthropic Watch analysis
What happened
The draft would allow a model to be blocked if its use posed an imminent risk of killing or seriously injuring more than 50 people, or causing more than US$1 billion in damage. A judge would have to approve the intervention, according to RTHK’s account of the proposal, first reported by Politico.
Anthropic says the bill would require frontier developers to test their most powerful models, disclose findings publicly, submit to independent evaluation and maintain a robust security programme. The report says the proposal would also override certain state laws, including California’s. Anthropic had argued in June that states should retain full authority in the absence of strong federal legislation.
Why it matters
This proposal would put real federal powers, and specific thresholds for using them, into the AI safety debate. It also raises a live question for states: whether a federal framework would provide a consistent floor, or displace protections they have already enacted.
The draft is circulating among lawmakers; the report does not say Congress has introduced or passed it. That distinction matters. A proposal can be ambitious on paper and still have a long, politically crowded journey ahead.
Our read
The notable move is the combination of developer testing and transparency duties with a possible government power to stop a dangerous deployment. Requiring judicial approval gives that power a defined check, though the thresholds and the meaning of “imminent risk” would matter enormously in practice. For now, treat this as Anthropic’s proposed blueprint, not US law.
What to watch
- Whether lawmakers formally introduce the draft or propose substantial changes.
- How the bill defines frontier developers, qualifying risks and the evidence needed for judicial approval.
- Which state laws the proposal would displace, and how state officials respond.
Discussion spark: Should a federal law be able to override state AI rules when a court finds an imminent danger, or should states retain the power to set stricter limits?
Sources and evidence
- Anthropic pushes own AI safety draft law in Congress – RTHK (17 October 2026, 00:00 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.