Anthropic has confirmed that it is circulating a proposed AI bill among U.S. lawmakers, according to AFP's report published by The Hindu. The report describes a proposal for federal oversight of the most powerful models, including testing requirements and a route for blocking dangerous deployments with a judge's approval.
This is a draft proposal described in reporting, not an enacted set of obligations. That distinction matters for anyone building a compliance plan around the announcement. The account explains what Anthropic is asking lawmakers to consider; it does not establish which language Congress would adopt or how a final law would operate.
A proposal focused on high-consequence deployments
AFP says Anthropic confirmed the circulation on October 16. According to the reported draft, the government could seek to block a model's release or use where there is an imminent risk of killing or seriously injuring more than 50 people, or causing more than $1 billion in damage. The report says that power would require judicial approval.
Those thresholds belong to AFP's description of the draft. They should not be read as a general definition of acceptable AI risk, or as instructions for determining whether an existing product complies with current law. The underlying legislative language and any subsequent changes would be needed for that kind of analysis.
Testing and outside evaluation are part of the pitch
An Anthropic spokesperson told AFP that the proposal would require frontier developers to test their most powerful models, disclose findings publicly, submit models to independent evaluation and maintain a security programme. The reported approach therefore involves preparation and scrutiny before a government attempt to stop a particular deployment.
For readers following AI safety, the concrete issue to watch is how those requirements would be specified. A public testing requirement still leaves questions about the scope of testing, the evidence released and the evaluator's access. The article reports the company's stated goals; it does not resolve those implementation details.
The state-law question remains consequential
The report also says the proposal would override certain state laws, including laws in California. That creates a second question alongside the proposed federal safeguards: what protections would remain in place while a federal framework is negotiated and implemented?
AFP includes a mixed outside assessment from Charlie Bullock of the Institute for Law & AI. He welcomed the proposal while describing it as weaker than the competing FRONTIER Act. His assessment is a reported opinion, rather than a settled comparison of final legislation.
What to follow after the announcement
The next useful evidence is an exact public bill text and a record of its legislative progress. Readers can then compare definitions, disclosure duties, enforcement procedures and the treatment of state authority without relying on a launch summary.
Our reading is that this proposal puts both safety obligations and limits on state authority into the same debate. Whether that produces stronger oversight depends on the eventual text and process. The immediate development is Anthropic's confirmed circulation of a draft, not the arrival of a new federal AI law.