Anthropic reportedly proposed its own AI-safety bill to the US Congress. It would require advanced-model risk assessments, incident reporting and whistleblower protection, while preventing independent evaluators’ compensation from depending on results. Imminent serious risks would be reported to the Commerce Department, with shutdown requests handled through specified federal courts. It also includes limits on other agencies’ shutdown authority and some state rules. The company confirmed the proposal, but legislative acceptance and enactment remain unresolved, so current obligations have not changed.