Anthropic representatives have quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security, Politico reported on October 16, 2026, citing three people with knowledge of the efforts. The striking part: the lab that builds Claude wrote the law that would govern it.
The pitch
Anthropic representatives brought a confidential draft bill directly to Capitol Hill lawmakers, according to Politico's October 16 reporting, citing three people with knowledge of the efforts. An Anthropic spokesperson told Reuters the bill represents proposals in the company's "Advanced AI Framework," released earlier this year and advocated with federal and state lawmakers since June. The framework would grant the government "the ability to block or deter dangerous deployments of the most powerful models" while requiring developers to test frontier models, be transparent about findings and get them independently evaluated. The Department of Commerce did not immediately respond to Reuters' request for comment.
What the draft would do
The trigger: catastrophic-scale thresholds only. 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. Routine harms — bias, misinformation, labor displacement — would not qualify.
The risk categories: three doors, all catastrophic. The qualifying risk must stem from one of:
- The creation or release of a biological weapon
- The model evading its developer's controls
- A cyberattack
Independent evaluators, registered with Commerce. Evaluators would register with the Department of Commerce and periodically evaluate each lab and its frontier models. Labs could pay the evaluators — provided payment is not tied to the evaluation's outcome. An evaluator finding "imminent enumerated risk" must report it to Commerce within 72 hours of completing its assessment.
One department, one court, final say with judges. Commerce could ask a federal judge to halt a model after determining risk, and a federal judge for the District Court for the District of Columbia could order the suspension. Crucially, the language explicitly prohibits any other US government department, agency or officer from suspending a model's development, deployment or internal use. Judges get the final say.
Why a lab wants federal guardrails now
Anthropic asking to be regulated looks paradoxical until you consider the alternative it fears. A patchwork of fifty state AI laws — each with different testing, disclosure and liability regimes — is the compliance nightmare frontier labs most want to avoid. A single federal framework with clear thresholds gives labs one rulebook, one regulator and, just as important, predictability for investors about what can and cannot trigger a government shutdown of a model.
There is also timing. Anthropic is reportedly preparing a US IPO that could value it at more than $2 trillion, with an expected listing as early as November — and New York City Comptroller Mark Levine has publicly urged the company to delay the offering until it addresses safety, governance and development-pace concerns. Arriving at the public markets with a self-designed safety framework already in lawmakers' hands is a way to convert a governance liability into a governance asset.
The regulatory-capture debate
The obvious criticism is regulatory capture in plain sight: the industry's leading safety-conscious lab is writing the rules that would bind it and its rivals. Critics argue the thresholds are set so high that they would almost never trigger, effectively giving labs a federal seal of approval while changing little day to day. The evaluator scheme, paid for by the labs themselves, invites the same auditor-capture dynamics that plagued credit-rating agencies.
Anthropic's counter-argument is that someone has to write the first draft, and a lab that has publicly warned about AI risks — including in CEO Dario Amodei's own essays urging the industry to slow down — is the most credible author. The company's position is that catastrophic-risk thresholds focus regulatory firepower where it matters, and that judicial review prevents arbitrary agency overreach. A single accountable department, in this telling, is better than a free-for-all of overlapping agency actions.
Both things can be true: the draft may be the most serious federal AI safety proposal yet floated, and it may also be designed to be the most comfortable one for the company that wrote it. Lawmakers — and the public — will have to decide whether to take the framework as a starting point or send Anthropic back with a blank page.
Why it matters for founders and investors
Compliance shape is now a frontier-AI business variable. If a framework like this becomes law, frontier-model developers will need budgeted, recurring relationships with registered independent evaluators, 72-hour incident-reporting machinery, and legal infrastructure for potential D.C. court proceedings. That is a fixed cost of building at the frontier — and a moat for whoever can afford it.
A new compliance industry is being born. Independent model evaluation — red-teaming, capability assessment, biosecurity review — is a service category the bill would make mandatory and fundable. Evaluator firms, audit tooling and transparency-reporting infrastructure are investable now, before the mandate lands.
The federal-versus-state question is the real fight. The draft's prohibition on other federal agencies suspending models signals Anthropic's preference for centralized, predictable authority. Whether states keep their own parallel AI safety regimes will determine whether AI startups face one rulebook or fifty-one. Investors underwriting frontier-AI companies should be tracking this fault line as closely as the technology itself.
Angeled.com connects founders building in this space with investors.
Read Next
Open Weights Are Not Open Business Models: Mistral Large 4 and the Licensing Trap
Courts Are Writing AI's Rules Now: Three Rulings Founders Should Read
Delhi High Court Orders 24-Hour Takedown of AI Deepfake Videos Targeting Chief Minister
EU Court Adviser Says AI Training Is Legal — but Gemini's Answers May Still Infringe Copyright
AI Regulation After the Midterms: What Founders Should Do Before the January Bills Arrive
DOJ Launches Super Intelligence Task Force to Police AI Misuse



