Anthropic’s Dario Amodei wants governments to have the power to block ‘dangerous’ AI systems

Anthropic’s CEO Dario Amodei has called for the U.S. government to adopt aggressive measures to regulate artificial intelligence, including the power to block deployment of high-risk systems. In a blog post, Amodei argued that AI models exceeding a specified compute threshold should trigger mandatory third-party audits before public release, with governments able to halt deployment if unacceptable risks—such as cybersecurity vulnerabilities or biological weapon capabilities—are identified. The proposal goes beyond President Trump’s June 2026 executive order, which encouraged voluntary model sharing with intelligence agencies for risk testing. Amodei’s framework demands independent auditors with veto authority, drawing parallels to regulations for aircraft, automobiles, and pharmaceuticals. He highlighted four key risk categories: cybersecurity flaws, biological weapon potential, accelerated dangerous research, and models surpassing human control. Anthropic has previously advocated for stricter AI controls, including its 2023 Responsible Scaling Policy, which introduced AI Safety Levels (ASLs) to manage risks as models advance. The company also supports legislation like California’s SB 53, which explores subsidies to mitigate AI-driven job displacement. Amodei warned that AI’s economic disruption could outlast past technological shifts, requiring policies to ensure equitable benefit-sharing. The blog post contrasts with Trump’s order by eliminating voluntary compliance, instead framing oversight as a mandatory safety requirement. Amodei emphasized that while AI drives economic growth, its risks demand proactive governance to prevent harm. His call for government intervention reflects growing concerns about unchecked AI development and its societal impact.