Anthropic has submitted a confidential draft of a federal bill to US lawmakers focused on the safety and security of frontier AI models. The proposal, reported on October 16, 2026, outlines conditions under which the federal government could intervene in the deployment of advanced systems.
The bill grants the Department of Commerce the authority to petition a federal judge to halt the deployment of a model if it poses an imminent catastrophic risk. Intervention is restricted to scenarios involving biological weapons, loss of model control, or cyberattacks.
Specific thresholds for government action include potential damages exceeding $1 billion or incidents causing death or serious injury to more than 50 people. Additionally, companies must report emergent risks to the Department of Commerce within 72 hours.
The proposal mandates independent testing and ongoing evaluation of advanced AI models. Evaluators must register with the Department of Commerce and cannot receive performance-based compensation. This framework draws inspiration from California’s SB 53, enacted in 2025.
