Anthropic Proposes Federal AI Safety Bill to Limit Government Intervention
- The proposed bill would restrict government intervention in advanced AI models, allowing action only for risks exceeding $1 billion in damage or potential fatalities of over 50 people.
- Developers would be required to maintain safety systems, publish transparency reports, and notify the Department of Commerce about critical incidents.
- Independent evaluators would audit AI models, with findings needing to be reported within 72 hours if imminent risks are identified.
- The draft seeks to roll back certain state-level regulations on public risk disclosures and model evaluations, raising concerns among some lawmakers.
- Other congressional proposals include the FRONTIER Act, which also aims to allow pauses in dangerous model deployments while preempting some state laws.
Anthropic’s proposal reflects ongoing debates in Congress regarding the balance between innovation and safety in AI development. The bill’s limitations on federal authority could significantly impact how advanced AI technologies are regulated across states.
With thresholds set for halting AI models at severe risks, this legislation could reshape compliance standards for developers amid growing scrutiny over AI safety measures. (Source)
