Anthropic, the developer of the Claude chatbot, has sent a draft bill to U.S. lawmakers aimed at regulating AI safety risks. This was reported by Politico and cited by Russian media. According to the startup's representatives, the document is based on the framework for creating advanced AI that the company has been promoting at the federal and regional levels in the U.S. since June 2026 through members of the U.S. Parliament. Anthropic believes that rules sensitive to both the industry and society should be considered by the U.S. Congress in the near future.
Context: Relationships with Authorities and Lobbying
Anthropic's relationship with U.S. authorities, according to reports, is complex: since March 2026, the company has reportedly been classified as a provider posing a national security threat. Nevertheless, the startup strives for a constructive dialogue with government representatives. Previously, Anthropic had not publicly lobbied for federal AI laws, limiting its work to the state of California, where it had already engaged in lobbying activities.
Key Conditions: Judge, Damage Threshold, and Risk Types
The central provision of the bill is to grant the government the power to block the release of an AI model only if the assumed and unavoidable damage from its release could total at least $1 billion, or lead to the death or serious injury of at least 50 people. Furthermore, only a federal judge of the U.S. District Court for the District of Columbia in Washington may impose a ban on the development and release of a model, with the decision based on independent audit results. Anthropic insists that no other U.S. agency or official should interfere in this process. The list of risks considered in the document includes the potential for creating biological weapons, loss of control over the model's distribution, and its use for cyberattacks. Such rhetoric, according to sources, is already present in the California AI regulation legislation passed last year.
Independent Audit and a 72-Hour Reporting Rule
A large portion of the document is dedicated to the mechanism of independent technological auditing. It proposes empowering independent inspectors registered with the U.S. Department of Commerce to conduct periodic audits of the startup and the models it develops. Anthropic is willing to pay for the auditors' work, while asserting that this will not affect the verification results, and expert fees should not depend on the outcome. Auditors are required to report to the Department of Commerce regarding the discovery of imminent major risks within 72 hours of the audit's completion. The agency then gains the right to seek a court injunction against further development of the dangerous model.
Legislative Context and Timelines
According to experts, Anthropic's initiative may preempt other legislative proposals currently promoted in the U.S. Parliament—specifically, rules for disclosing information about advanced AI risks, audit procedures, and publishing reports on critical incidents. A bill formed from fragmented state initiatives, as noted, is harder to approve than uniform federal rules. In addition to regular audit requirements, the authors believe that bills should provide whistleblower protections for those ready to report on unreliable developers, as well as procedures for publishing incident reports. That said, no recorded AI incident to date has resulted in damages exceeding $1 billion or mass casualties, making the proposed thresholds effectively preemptive.
