Dario Amodei, CEO of Anthropic, published a new policy essay on June 10, 2026, urging mandatory third-party safety testing for frontier AI models. He proposed a regime modeled on the FAA, requiring pre-release audits in cybersecurity, biological weapons, loss of control, and automated AI research. Governments should be empowered to block systems that fail safety audits. The shift from transparency-based disclosure laws to binding rules reflects the accelerating pace of AI progress. Amodei highlighted cyber risks as fully materialized, with Anthropic's Claude Mythos Preview solving 73% of expert-level challenges. The essay warns of potential disruptions to financial sectors and critical infrastructure, echoing concerns about DeFi protocol vulnerabilities. On economic impacts, Amodei proposed wage insurance and workforce training, with universal basic income as a last resort. He also called for a ban on fully autonomous weapons in domestic law enforcement and tighter data broker regulations. A coalition of democracies should control chip exports, citing US bills MATCH and OVERWATCH.

Key facts

• Anthropic CEO Dario Amodei calls for mandatory third-party audits of frontier AI models.
• Models failing safety audits could be blocked or reversed by governments.
• Claude Mythos Preview solved 73% of expert-level cyber challenges.
• Essay warns AI disruption to financial sector and critical infrastructure.
• Proposes wage insurance and workforce training for job displacement.
• Calls for ban on fully autonomous weapons in domestic law enforcement.

KeyAudit data perspective