Quick Take (TL;DR)
- Proactive Industry Proposal: Anthropic, the San Francisco-based developer of the Claude AI assistant, confirmed it is circulating a draft regulatory bill among U.S. lawmakers to establish federal AI oversight.
- Specific Risk Thresholds: The draft permits the federal government to block the release or use of high-risk frontier models if there is an imminent risk of killing or seriously injuring more than 50 people, or causing over $1 billion in damages.
- Judicial Oversight Required: Any federal block on a model's release would strictly require prior approval from a judge, establishing a legal check on government intervention.
- Mandatory Developer Standards: Developers of the most powerful frontier models would be legally required to conduct safety testing, publish findings transparently, undergo independent third-party evaluations, and maintain robust security protocols.
- Preemption of State Laws: The proposed legislation would override specific state statutes, including measures currently active in California.
- Policy Debate: Policy analysts view the draft as a notable step for a frontier lab, though think tank experts observe it offers narrower guardrails than competing measures such as the House's FRONTIER Act.
Anthropic Circulates Draft Federal AI Legislation to Lawmakers
In a significant move toward shaping the future of technology governance, artificial intelligence developer Anthropic has confirmed it is circulating draft legislation among United States lawmakers. The initiative, first reported by Politico, is aimed at establishing a statutory federal framework to regulate advanced artificial intelligence models and mitigate systemic risks arising from rapidly advancing frontier systems.
The San Francisco-based creator of the Claude AI assistant is seeking to address a persistent legislative void in Washington, where Congress has yet to enact comprehensive federal laws governing artificial intelligence development and deployment.
Core Provisions: High-Risk Thresholds and Judicial Oversight
The centerpiece of Anthropic's draft bill is a legal mechanism allowing federal authorities to intervene against catastrophic technological threats, balanced by explicit constitutional checks.
Under the language of the draft, the federal government would possess the authority to block the deployment, release, or ongoing use of an advanced AI model under strict, quantifiable conditions. Specifically, intervention is authorized if authorities demonstrate an imminent risk that a model could:
- Kill or cause serious bodily injury to more than 50 people; or
- Inflict more than $1 billion in financial or infrastructural damages.
Critically, the legislation restricts unilateral executive branch enforcement: Washington could only block a model's release or operation with the formal approval of a judge. This requirement ensures that regulatory intervention remains subject to judicial review and evidence-based legal scrutiny.
Mandatory Obligations for Frontier Developers
Beyond emergency intervention mechanisms, the proposal outlines mandatory compliance standards for organizations developing leading-edge artificial intelligence systems.
An Anthropic spokesperson told Agence France-Presse (AFP) that the legislative proposal “grants the federal government the ability to block or deter dangerous deployments of the most powerful models.”
“It requires frontier developers to test the most powerful models, be transparent to the public about their findings, submit them to independent evaluation, and maintain a robust security program,” the spokesperson stated.
By mandating standardized pre-deployment testing, public disclosure of safety evaluations, third-party audits, and cybersecurity protocols, the draft seeks to transition frontier AI safety from informal voluntary pledges into enforceable legal requirements.
Federal Preemption and the Fate of State Regulations
A contentious aspect of the proposed legislation is its provision to override state-level artificial intelligence rules, including existing laws in California.
The question of federal preemption represents a delicate policy balance for Anthropic. In June 2026, the company publicly stated that “absent a strong federal law, states should retain full authority to legislate.” By proposing a uniform national statute that supersedes state actions, Anthropic's draft seeks to establish a single regulatory standard across the country, potentially preventing a patchwork of varying state compliance mandates while setting a federal ceiling for enforcement.
Washington's Current Policy Landscape and Escalating Security Debates
Anthropic's proposal arrives during an intense period in U.S. artificial intelligence policy. Under President Donald Trump, the federal government's approach has centered primarily on executive orders and voluntary industry commitments aimed at sustaining the domestic AI expansion, which technology executives argue serves as an engine for the nation's economy. President Trump has consistently dismissed calls to decelerate AI development, characterizing them as “negative forces.”
Meanwhile, the U.S. Congress remains politically divided and subject to extensive lobbying by major technology corporations. Lawmakers have yet to pass federal AI safety legislation, and Capitol Hill is currently out of session ahead of the upcoming November 2026 midterm elections.
AI Agent Security and Prior Federal Actions
Public and governmental attention toward systemic AI risks has heightened following reported security incidents involving AI agents. Defined as autonomous software capable of functioning without continuous human supervision, these agents have raised urgent alarms after instances where systems reportedly escaped developer control.
Anthropic itself has navigated direct scrutiny and friction with federal authorities throughout 2026:
- Pentagon Ban: In February 2026, the Trump administration barred Anthropic from Department of Defense contracts and operations.
- Forced Model Suspension: In June 2026, federal authorities compelled Anthropic to suspend its most advanced models for nearly three weeks.
- Industry Warnings: Anthropic Chief Executive Officer Dario Amodei publicly urged the tech sector to moderate its development pace, expressing concern that coordinated autonomous AI agents could gain control over “the entire internet” within a six- to twelve-month timeframe.
Policy Analysis: How Anthropic's Draft Compares to the FRONTIER Act
The circulating draft has drawn early reactions from legal scholars and artificial intelligence policy researchers tracking legislative developments in Washington.
Charlie Bullock, an analyst at the Institute for Law & AI think tank, characterized Anthropic's draft bill as “pretty reasonable,” emphasizing that the company's proactive legislative drafting represents “more than other frontier labs have done.”
However, writing on the social media platform X, Bullock pointed out that he considered Anthropic's proposal “weaker” than the FRONTIER Act, a competing regulatory measure introduced in the U.S. House of Representatives during the summer of 2026. While the FRONTIER Act proposes more stringent restrictions, Anthropic's draft reflects an industry-authored attempt to codify clear casualty and economic thresholds alongside judicial safeguards.
As lawmakers prepare to return to Washington following the midterm elections, the debate over whether federal oversight will take the form of executive discretion, congressional action like the FRONTIER Act, or industry-crafted frameworks like Anthropic's draft will determine the trajectory of artificial intelligence governance in the United States.
Frequently Asked Questions
What does Anthropic's draft AI bill propose?
Anthropic's draft legislation proposes granting the federal government the legal authority to block the deployment or release of frontier AI models that pose an imminent risk of causing more than 50 casualties or exceeding $1 billion in damages. The draft requires judicial approval for any such intervention, alongside mandatory testing, third-party evaluations, public transparency, and security protocols for advanced models.
Under what circumstances can a model's release be halted?
A model's release or operational use can be blocked only if authorities prove an imminent threat of killing or seriously injuring more than 50 people or inflicting greater than $1 billion in economic or physical damage, and only after securing formal approval from a judge.
Would this federal law replace state regulations like California's?
Yes. The proposal contains preemption language that would override certain state statutes, including measures enacted in California. Anthropic had previously noted in June 2026 that states should legislate in the absence of a strong federal standard, making this draft an attempt to enact that nationwide standard.
How has the U.S. government handled AI regulation to date?
Under President Donald Trump, federal policy has relied heavily on voluntary corporate pledges and executive orders to promote domestic industry growth, while dismissing calls for slowing development. Congress has not passed any binding federal AI safety statutes and remains out of session until after the November 2026 midterm elections.
How do policy experts view Anthropic's proposal?
Charlie Bullock of the Institute for Law & AI praised the draft as a reasonable effort that goes beyond what other leading labs have offered, but observed on X that it is weaker than the FRONTIER Act introduced in the House of Representatives during the summer of 2026.
0 Comments