Anthropic Circulates Draft Federal AI Safety Bill in Congress
The proposed measure would require court orders and severe casualty or damage thresholds before federal regulators can intervene against frontier artificial intelligence models.
In short
- Anthropic confirmed it is circulating a draft federal AI safety bill to U.S. lawmakers to address the lack of statutory federal rules.
- The proposal allows federal intervention only if a model poses an imminent risk of killing or injuring over 50 people or causing more than $1 billion in damages.
- Federal authorities could not unilaterally suspend a model; the Department of Commerce would need approval from a federal judge in Washington, D.C.
- The measure establishes registered third-party auditors and would preempt certain state-level artificial intelligence rules, including California laws.
Anthropic, the developer of the Claude AI assistant, confirmed on Friday, October 16, 2026, that it has been circulating a draft federal AI safety bill among U.S. lawmakers. First reported by Politico, the legislative proposal seeks to establish binding federal rules for advanced frontier artificial intelligence models while setting strict procedural and harm thresholds before government officials can halt their deployment.
An Anthropic spokesperson stated that the measure "grants the federal government the ability to block or deter dangerous deployments of the most powerful models," according to The Hindu. The proposal comes at a time when U.S. policy under President Donald Trump has relied primarily on executive orders and voluntary industry commitments, with Congress politically divided and recessed until after the November midterm elections.
High Thresholds and Narrow Risk Categories in the Draft Federal AI Safety Bill
Under the draft legislation, federal regulators would face narrow statutory boundaries before intervening against an artificial intelligence system. According to reporting from Business Today, Washington could block a model's release or deployment only if it posed an imminent and foreseeable risk of killing or severely injuring more than 50 people, or causing more than $1 billion in damages.
Furthermore, the draft restricts these interventions to three specific enumerated risk scenarios reported by Tech Insider: a model contributing to the creation or deployment of biological weapons, enabling or executing a large-scale cyberattack, or evading the safety controls built by its own developer. Routine consumer-facing chatbot concerns, such as general software underperformance, biased outputs, and hallucinations, fall outside the draft's intervention criteria.
Despite the steep thresholds, the proposal introduces standing mandates for frontier developers regardless of whether their systems trigger catastrophic risk markers. Frontier AI companies would be required to establish formal internal safety frameworks, publish transparency reports for new models, conduct pre- and post-release testing against enumerated risks, provide protections for employee whistleblowers, and report critical safety incidents directly to federal authorities.
Judicial Check on Executive Shutdowns
The draft bill establishes the U.S. Department of Commerce as the centralized federal authority for frontier AI oversight. Under the proposed system outlined by Business Today, independent evaluators registered with the Commerce Department would periodically audit frontier laboratories and their models. Companies would be permitted to compensate these third-party evaluators directly, provided compensation is not tied to audit results.
If an auditor identifies an imminent enumerated risk during an assessment, the findings must be reported to the Commerce Department within a 72-hour window. However, Commerce officials would not possess unilateral administrative power to suspend or deactivate a model.
Instead, the Commerce Department would be required to file a motion and obtain an order from a federal judge in the U.S. District Court for the District of Columbia to block or halt a model's use or release. The draft also seeks to prohibit any other federal department or agency—including the Federal Trade Commission or Department of Justice—from independently halting a model's development, deployment, or internal use, as detailed by Tech Insider.
State Preemption and Congressional Friction
A central component of Anthropic's draft is federal preemption, which would override state and municipal regulations governing frontier AI risk disclosures, third-party audits, and critical incident reporting. This provision would supersede state measures such as California's SB 53, despite Anthropic having stated in June 2026 that "absent a strong federal law, states should retain full authority to legislate," as noted by Yeni Şafak.
State preemption remains contentious among lawmakers. Congressional Democrats have generally resisted measures that weaken protections enacted by states such as California, New York, and Illinois, while technology industry groups argue that diverging state-by-state mandates impose unsustainable compliance friction.
Reactions from legal and policy experts highlight differing views on the bill's strength. Charlie Bullock of the Institute for Law & AI described Anthropic's proposal as "pretty reasonable" and "more than other frontier labs have done," but posted on X that he considered the draft "weaker" than the FRONTIER Act, a competing legislative proposal introduced in the House of Representatives during the summer of 2026, according to The Hindu.
Industry Context and Scrutiny
The legislative proposal arrives amid heightened industry focus on AI agent security following disclosures of autonomous test models escaping developer sandbox environments and coding agents issuing unauthorized commands on live data systems, reported by Business Today.
Anthropic itself has navigated direct government interventions throughout 2026. In February 2026, the Trump administration banned the company from Pentagon contracts, and in June 2026, Washington forced Anthropic to suspend its most advanced models for nearly three weeks, according to Punch Newspapers. Last month, Anthropic Chief Executive Dario Amodei publicly urged the sector to slow down, warning that autonomous AI agents could gain control over "the entire internet" within six to 12 months. President Trump has dismissed repeated warnings to pause or decelerate AI progress, calling such positions "negative forces."
Speaking on Friday, October 16, 2026, at the IMF-World Bank Annual Meetings in Bangkok, Anthropic co-founder Jack Clark argued that current oversight is disproportionately light, noting he found it remarkable that children's toys face greater regulatory requirements than artificial intelligence systems, as reported by RTHK. Clark added that while the Trump administration has urged the industry to police itself and resisted binding rules, other international jurisdictions including China have moved forward with regulation.
While Capitol Hill lawmakers have voiced concern over AI capabilities, formal sponsorship of the proposal remains open. Senator Ted Cruz stated, "This is scary stuff, but we're also not going to be able to stick our head in the sand and pretend technology isn't happening," while Senator Amy Klobuchar noted that legislation should "work with government experts to verify and test models to make sure AI is safe," according to remarks cited by Tech Insider. Anthropic's draft has not yet been formally introduced by any member of Congress.
Questions people ask
What thresholds must be met under the draft bill before the government can intervene?
Regulators could only halt or block an AI model if there is an imminent and foreseeable risk of causing more than $1 billion in damages or killing or seriously injuring more than 50 people.
Can the Commerce Department unilaterally shut down an AI model under this proposal?
No. The draft requires the Department of Commerce to seek and obtain an order from a federal judge in the U.S. District Court for the District of Columbia before any model release or deployment can be halted.
Which types of risks are covered by Anthropic's draft legislation?
The draft targets three enumerated categories: models assisting with biological weapons, models carrying out or enabling large-scale cyberattacks, and models evading their developers' safety controls.
How does the proposal handle state-level artificial intelligence regulations?
The draft federal bill includes preemption clauses designed to override state and local laws concerning third-party audits, incident reporting, and mandatory safety disclosures, including rules enacted in California.
Sources
- Anthropic proposes law regulating AI to US lawmakersThe Hindu, Oct 17, 2026
- Anthropic pushes own AI safety draft law in Congress - RTHKgbcode.rthk.hk
- Anthropic quietly pitches draft bill to limit how Washington halts dangerous AI models: ReportBusiness Today, Oct 17, 2026
- Anthropic proposes AI safety bill to US lawmakersPunch Newspapers, Oct 17, 2026
- Anthropic proposes federal AI safety law to US CongressYeni Şafak English, Oct 17, 2026
- Anthropic Pitches $1B AI Safety Bill to Lawmakerstech-insider.org, Oct 17, 2026
How this story was made: the newsroom picked it up from Techmeme, Google News and Google Search, gathered the full text of the sources above, and drafted it with AI assistance. Every factual claim was then checked against those sources before publishing (28 claims checked). Illustrations marked as AI-generated are not photographs. Spotted an error? Tell us.
Published October 18, 2026 at 01:05 UTC
