Dario Amodei calls for government power to block unsafe AI releases

Anthropic CEO Dario Amodei has called for stronger AI oversight, including government authority to prevent the release of dangerous systems. His proposal would move beyond requiring companies to disclose their safety practices and introduce mandatory outside testing.

The proposed tests would cover cyberattacks, biological weapons, systems escaping human control and automated research that could intensify those dangers. Government intervention would be limited to those risks, with protections against politically motivated or arbitrary decisions. Developers would also need stronger security and prompt reporting of serious incidents.

Amodei argued that transparency requirements were an appropriate earlier step, but that emerging dangers now justify enforceable safeguards. Anthropic has published separate frameworks addressing advanced AI oversight and potential disruption to employment. These are policy recommendations, and the proposed testing system would require government action to become binding.