Anthropic's draft AI bill, as reported, next to its June framework

There is nothing to switch on or migrate here. If you build on Claude, this is a read on how Anthropic wants frontier AI regulated in the US, and on whether a federal law could replace state rules like California's SB 53. Politico reported on Friday that Anthropic has been quietly pitching lawmakers a draft bill. An Anthropic spokesperson told Reuters the bill represents proposals outlined in Anthropic's Advanced AI Framework, which the company published in June. This is reported, not confirmed: Anthropic has not posted the bill, and none of its X accounts has mentioned it.

What changed#

Politico's Owen Dahlkamp reported on Oct 16 that Anthropic representatives had pitched lawmakers on Capitol Hill a draft bill on frontier AI safety and security, citing three people with knowledge of the effort. In Politico's account (paraphrased here, since we could not open the page itself), the draft would narrow when the government could block an advanced model's release, and it would preempt some state AI safety laws.

According to Politico, as summarized by Reuters:

• The government could block a model's release or use only if it poses "a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people."
• The risk must come from a biological weapon, the model evading its developer's controls, or a cyberattack.
• Independent evaluators would register with the Commerce Department and periodically evaluate labs and their frontier models. Labs could pay them, as long as payment is not tied to the outcome.
• Evaluators who find "imminent enumerated risk" must report it to Commerce within 72 hours.
• Commerce could ask a federal judge in the District of Columbia to halt a model. Reuters says the language "explicitly prohibits any other US government department, agency, or officer" from suspending development, deployment or internal use.

Anthropic's side, per the spokesperson quoted by Reuters: the bill would give the government the "ability to block or deter dangerous deployments of the most powerful models," and would require developers to test frontier models, be transparent about findings, and get them independently evaluated.

Before vs after: the June framework and the draft as reported#

Anthropic's policy page also says it does "not believe Congress should preempt state law unless it enacts a federal law that is at least as strong as the framework." Whether this draft clears that bar is the core fight. Anthropic's spokesperson frames the bill as the framework's proposals. Critics say it falls short (see below).

Why it matters#

For most teams building on Claude, the obligations would not land on you. Both the framework and the reporting aim at the developers of the largest models. Two parts could still reach you. First, preemption: AI Policy Landscape, which summarized a leaked copy, says the draft preempts state "disclosure, third-party evaluation, and incident-reporting obligations on any AI developer," not only the largest ones. Second, the halt rule: a model you depend on could only be stopped through one court process, on a high and narrow bar. Whether you see that as stability or as a weak safeguard is the debate now playing out.

How to read it yourself#

1. Read Anthropic's own words first: the Advanced AI Framework PDF, especially "Preemption" and "Possible approaches to establishing authority to block or deter unsafe models."
2. Read the Politico report (it may be paywalled) or the Reuters summary.
3. Treat any copy of the bill text online as a leaked draft. Anthropic has not published it.
4. To compare the two documents side by side, paste both into Claude with a prompt like this one. The prompt is ours and we have not run it:

Below are two documents: (A) Anthropic's June 2026 Advanced AI Framework and (B) a draft bill text.
For each of these topics, quote the exact passage from A and from B, then say in one sentence whether B is stricter, looser, or the same as A:
1. which developers are covered
2. who can stop a model, and the legal standard
3. which risks count
4. independent evaluators: access, payment, reporting deadlines
5. preemption of state law
If a document says nothing on a topic, write "not addressed". Do not infer missing text.

Good to know#

• Reported, not published. Every bill detail above comes from Politico's reporting on a confidential draft it obtained, relayed by Reuters, or from third-party readings of a leaked copy. Anthropic's on-record comment, per Reuters, is that the bill represents proposals outlined in its framework.
• Reactions are sharply split. In a post, Zvi Mowshowitz called it "unacceptable" and wrote that it is "similar to the FRONTIER Act, with stronger preemption of state laws, stronger transparency provisions and weaker enforcement provisions." Charlie Bullock, quoted in the same post, said Anthropic "deserves some credit for actually proposing a pretty reasonable AI safety bill," but that it "is weaker than the Trahan-Obernolte FRONTIER Act in some key ways."
• No sponsor yet. AI Policy Landscape lists the draft with "no congressional sponsor."
• Commerce had no comment. Reuters said the Department of Commerce "did not immediately respond" to a request for comment.

What to watch: whether a member of Congress introduces the bill, which would put an official text on congress.gov; whether Anthropic posts the text or a response on anthropic.com/news or @AnthropicAI; and whether the final language still preempts state disclosure and incident-reporting rules.