Anthropic CEO calls for government power to block risky AI

Government authority over AI deployment

Anthropic CEO Dario Amodei has outlined a sweeping regulatory vision that would grant governments legal authority to block or reverse AI deployments deemed dangerous to public safety. In an essay published Wednesday, Amodei argues that frontier AI models should undergo mandatory testing for cybersecurity risks, biological weapons potential, loss of control scenarios, and automated research and development capabilities before release.

The proposal, detailed by Axios, represents one of the most comprehensive regulatory frameworks yet proposed by a major AI company leader. Amodei explicitly compares AI regulation to existing safety regimes for cars, airplanes, and pharmaceuticals—industries where products must meet rigorous standards before reaching the public.

Why it matters

Amodei's framework goes significantly beyond current policy discussions in Washington and sets up potential conflict within the AI industry. Critics are likely to argue that Anthropic is advocating for strict rules that could cement its position against smaller competitors, or that the company is using catastrophic scenarios as a marketing strategy. The timing is notable: the proposals follow the release of Anthropic's Mythos model and suggest the company sees emerging risks on the near-term horizon.

Economic disruption measures

Beyond safety regulation, Amodei's essay addresses economic displacement from AI with proposals including wage insurance programs, retention tax incentives for employers, expanded social safety nets, and potentially universal basic income or universal capital accounts. He argues that AI could produce labor market disruptions larger and more lasting than previous technological shifts, and calls for improved data collection on AI-related job losses.

Amodei suggests that public resistance to data center construction stems less from local concerns than from broader economic anxiety about AI's impact on work and livelihoods.

Regulatory speed and scope

The Anthropic CEO frames the core challenge as a mismatch between AI's rapid advancement and the slow pace of policy development. He argues that existing transparency legislation falls short and that "more serious and binding regulation" is necessary. His framework would expand the Trump administration's AI executive order to include mandatory risk testing.

Looking ahead, Amodei indicates that even more aggressive regulation might become necessary if AI systems pose greater threats. He also calls for regulatory reform at agencies like the FDA to accelerate approval of AI-discovered drugs, and advocates for banning fully autonomous weapons domestically while maintaining democratic leadership on AI governance internationally.

Despite the emphasis on risks and regulation, Amodei maintains optimism about finding workable solutions. He notes that many AI policy issues—from job displacement to model testing to chip export controls—have potential for bipartisan support.

These details were first reported by Axios.