Why Anthropic CEO Dario Amodei thinks a 10-year AI regulation freeze is dangerous

Dario Amodei, Chief Executive of AI firm Anthropic, has criticised a Republican proposal to halt state-level regulation of artificial intelligence for ten years, calling the measure "far too blunt" in a sharply worded opinion piece published inThe New York Times.

The proposal, reportedly embedded within former President Donald Trump’s revived tax cut bill, seeks to prevent individual US states from enforcing their own AI regulations. This move, backers say, is intended to create a unified national framework. However, it has faced significant pushback, including from a bipartisan coalition of state attorneys general, who have already enacted safeguards against high-risk AI use.

Amodei, whose company is backed by Amazon, argued that the rapid pace of AI advancement demands a more agile and balanced approach. “A 10-year moratorium is far too blunt an instrument. AI is advancing too head-spinningly fast,” he wrote. “Without a clear plan for a federal response, a moratorium would give us the worst of both worlds — no ability for states to act, and no national policy as a backstop.”

Instead, Amodei called for a coordinated effort between the White House and Congress to develop a federal transparency standard. Such a framework, he suggested, would compel AI developers to openly disclose their testing methods, risk mitigation strategies, and national security considerations prior to releasing advanced models.

Anthropic already shares these details publicly, Amodei noted, and rivals OpenAI and Google DeepMind have adopted similar practices. However, he cautioned that voluntary transparency may not be sustainable as AI systems become more powerful and commercial stakes rise. “Legislative incentives may become necessary to ensure this openness continues,” he said.

Amodei’s intervention adds weight to calls for a federal framework that preserves public safety while ensuring that innovation is not stifled, a balance that may prove increasingly difficult to strike as AI tools become more sophisticated and deeply embedded in daily life.

(With inputs from Reuters)