Deep dive: Anthropic's Leaked AI Bill and the Fight Over Preemption

Why today

On October 16, Politico reported that Anthropic had been quietly showing lawmakers on Capitol Hill a draft federal bill to regulate frontier AI. Politico had the confidential draft. Anthropic has not published it. The reaction has kept building since then. On October 17, Zvi Mowshowitz, who writes the newsletter Don't Worry About the Vase, called the leaked draft unacceptable and troubling. California State Senator Scott Wiener said it falls far short of protecting public health and safety. Anthropic has partly retreated. Its head of federal affairs, Anthony Cimino, said the preemption wording was "not precise enough" and would be updated. He also said the proposal is meant to be "the strongest on the table." So this is the moment to ask a structural question. Suppose a frontier lab helps write the federal rules, and those rules override the states. Who does that lock in, and who gets locked out?

Plainly

Here is the draft as Politico and other outlets describe it. The Commerce Department would register independent evaluators. Those evaluators would periodically assess frontier developers and their models. Developers could pay their own evaluators, as long as the payment is not tied to the result. Evaluators would have to tell Commerce within 72 hours if they find an imminent risk. The bill also requires risk reports and whistleblower protections. Zvi calls its security requirements strong. The fight is over two other parts. The first is the bar for stopping a model. The government could step in only when a model poses a foreseeable and imminent risk of more than $1 billion in damage, or of killing or seriously injuring more than 50 people. The risk would also have to come from a narrow set of scenarios: bioweapons, a massive cyberattack, a model evading its developer's safety controls, or research that speeds those up. Only a federal judge in Washington could order a suspension. The second part is preemption. That means a federal law displacing state law. The draft would override California's SB 53, New York's RAISE Act and Illinois' SB 315. One way to picture it is building codes. Today, states write their own fire codes for one specific kind of tall building. Under this draft, Washington writes the only code. Only Congress can change it. Inspectors are hired and paid by the builder, and no one licenses them. A city that sees smoke cannot close the building. It has to ask a federal judge, and the judge must weigh what closing would cost the owner. That picture also shows what the draft does not appear to cover. It is aimed at frontier developers. The reporting does not say who counts as one. A seed-stage company building on top of a model is probably not the regulated party. It may still feel the effects, through its suppliers and through the market for audits.

What the evidence shows

Start with what is established. Politico reported the draft from a confidential copy and from three people familiar with Anthropic's legislative work. Anthropic has not denied it. Cimino publicly said the company circulated the ideas to get early feedback from policymakers. Cimino later posted on X that Anthropic would update its preemption language to better match its own published framework. He also said the company would address the standard for blocking models, risk thresholds and requirements for evaluators. So the draft is real. Anthropic stands behind its direction, but not its wording. Several provisions are confirmed by more than one outlet. These include the $1 billion and 50-casualty thresholds, the federal court in Washington, the evaluators paid by labs, and the list of state laws it would preempt. Cimino says the harm thresholds were pulled directly from California law. That holds up, with a catch that Wiener points out. SB 53 uses those thresholds to trigger disclosure and assessment. The draft uses them to decide when a model can be stopped. Next come the legal readings, which are analysis, not fact. The draft itself tells the court to weigh the harm a blocking order would do to the developer, its users and the public. In his reading, they are not told to weigh the harm of not blocking. Charlie Bullock of the Institute for Law and AI warned in a post on X, quoted by Transformer, of a race to the bottom among auditors. In this reading, they would compete to offer the least bothersome audits. Nathan Calvin, as Zvi quotes him, argues that passing the bill would leave us less protected than existing law does. These are serious readings by people who know the field. They are also from safety advocates, and they are interpretations of a text the public cannot read. Then there is what nobody has shown. The full text is not public. Who counts as covered is not stated in any coverage we found. Neither is any compute or revenue cutoff. The revised draft and the promised framework for evaluators have not appeared. Any claim about how this affects small builders is, for now, inference.

How people read it

The same facts support several readings. The safety reading, held by Zvi, Calvin and Wiener, comes from people who take catastrophic risk seriously. In this view the draft is weaker than the bipartisan FRONTIER Act from Representatives Obernolte and Trahan. They argue it removes state backstops that work today and freezes weak federal rules that only Congress can tighten. Zvi's sharpest line is that Anthropic seems "more afraid that the government might enforce rules against Anthropic" than worried about the government's power to enforce rules against everyone. The market reading is the strongest case for the draft. A single federal standard is cheaper to comply with than fifty. Clear thresholds and a single court cut legal uncertainty for every developer. One outlet reports that child safety, consumer protection and civil rights laws are carved out of preemption. If so, states keep their role on everyday harms. The realist reading is more skeptical. The scenarios that would justify a shutdown are narrow, and none has happened yet. On that view, how blocking works matters less than who writes the standards for evaluators and who pays for them. That is where an incumbent's advantage would build up over time. The policy reading points to a shift in roles. Earlier, Anthropic endorsed SB 53 and similar state laws. It is now proposing to replace them. That doesn't prove bad faith. It does mean a lab's policy position is something we should track over time, not take as fixed.

What it means for builders and investors

Our read is that the label of incumbent capture fits the evaluation market more than it fits the founders. Frontier-level obligations will almost certainly fall on a few large labs. Most seed and Series A companies will not be covered directly. The indirect effects are where it gets real. One is evaluators who are paid by labs and not licensed. Another is standards that only Congress can update. A third is state regimes that disappear. Together they decide what assurance means for every model your product sits on. For startups selling evaluation, red-teaming or AI assurance, this is close to an existential question. A weak federal floor could create a big market for cheap compliance, but not for rigorous testing. For application companies, the question is quieter. Their enterprise customers may now point to SB 53 disclosures from their model providers. Those could vanish or change shape. We hold all of this loosely, because the text is not public and a revision has been promised. One question worth exploring with an assurance startup is this: if the federal standard were the floor in this draft, would the main buyers still pay for your product? Or does your revenue depend on state law that could be preempted?

What to watch

Three things would settle this. First is Anthropic's revised preemption language. Cimino promised it, but it has not appeared. When it does, the test is simple. Does it still stop states from blocking a model in an emergency, or only from setting conflicting disclosure rules? Second is the evaluator framework. Licensing, conflict-of-interest rules and a say for Commerce over standards would answer most of the race-to-the-bottom worry. Third is whether any member of Congress introduces the draft, or folds it into the FRONTIER Act. A bill text with a sponsor is the moment this becomes law in the making rather than a lobbying document. None of these has a public date yet.

Steven's take

Company’s becoming governments is a big part of sci-fi lore - and here we are living in the future.