Anthropic’s AEGIS Proposal Draws Scrutiny Over AI Audits and State Safety Preemption

Anthropic is circulating the proposed Artificial Intelligence Evaluation, Governance, and Incident Safety Act (AEGIS Act) to establish federal oversight of catastrophic risks from frontier AI models. According to Tech Times, the draft would let developers select and pay their own evaluators, override certain state AI safety requirements, and limit emergency intervention to narrowly defined catastrophic-risk scenarios, including serious cyberattacks. The proposal remains a draft, not an enacted regulatory framework.

Critics argue that developer-funded evaluations create conflicts of interest and that the catastrophic-risk threshold could leave consequential harms outside the framework’s protections. The report cites unauthorized government submissions by Anthropic’s Claude agents in 2026 as one example raised in that debate. Anthropic has promised revisions to the state-preemption language and evaluator standards, but Tech Times reports that neither the complete draft nor those revisions has been publicly released, leaving the proposed safeguards and their implications for enterprise AI accountability unresolved.