Anthropic Revises Draft Federal AI Safety Bill
The company aims to clarify preemption rules as it seeks to define federal oversight for high-impact artificial intelligence models.
Updated on Oct. 21, 2026 in Artificial Intelligence
Live Poll
Should federal AI safety laws take precedence over stricter state-level regulations?
Anthropic is revising a draft AI safety bill previously circulated on Capitol Hill after public criticism of its preemption language. The proposal outlines federal oversight mechanisms, including a process for the Commerce Department to petition courts to block AI models based on specific harm thresholds.
Why it matters
The draft represents a significant attempt by a major AI lab to define federal regulatory boundaries, including thresholds for government intervention that would override state-level disclosure and reporting requirements. It arrives as lawmakers continue to debate the legislative framework for AI governance following the introduction of the FRONTIER Act earlier this year.
The draft mandates that AI developers adhere to a $1 billion damage threshold or a 50-person injury limit to avoid government-led judicial model blocks. It also introduces a 72-hour reporting deadline for imminent AI risks, a timeline far stricter than current industry standard reporting windows.
The players
Anthropic
An AI research lab and developer of large language models focused on constitutional AI and safety-aligned deployment.
Anthony Cimino
The head of policy at Anthropic who oversees the organization's legislative and regulatory engagement strategies.
The details
The proposed bill establishes a federal oversight process where the Commerce Department must petition a federal court to halt the operation of an AI model. Under this framework, third-party evaluators must register with the Commerce Department to conduct safety checks. The bill also seeks to preempt state laws concerning AI public disclosure and reporting, a move Anthropic officials recently admitted requires more precise legislative phrasing.
Timeline
July 2026: The FRONTIER Act is introduced in Congress.
October 16, 2026: Politico publishes the confidential draft bill.
October 17, 2026: Anthony Cimino announces the bill will be rewritten.
The Tech Race
This draft bill follows the July 2026 introduction of the FRONTIER Act, marking a competitive effort to shape federal AI oversight standards. It arrives as the industry navigates a post-2026 environment where the current administration has already demonstrated a willingness to shut down specific AI models.
AI developers and companies operating in the U.S. should monitor these thresholds, as they establish the legal triggers for mandatory federal reporting and potential model suspension. If enacted, this legislation would simplify compliance for firms by replacing patchwork state requirements with a unified federal standard.
The takeaway
The pivot in the bill's drafting highlights the ongoing tension between lab-led voluntary safety standards and the emerging push for federal legislative control. Readers should watch for the forthcoming publication of updated proposals regarding the requirements and qualifications for the independent evaluators mentioned in the draft.
Further reading
For broader context on federal policy developments, visit Artificial Intelligence.
Live Poll
Should federal AI safety laws take precedence over stricter state-level regulations?
