Dario Amodei (다리오 아모데이), CEO of Anthropic, said on June 10 that it is time to shift AI regulation from transparency disclosure requirements to binding, enforceable rules.

Amodei made the argument in a public essay titled "Policy on the AI Exponential," writing that AI risks have now clearly become reality. He said Anthropic had supported transparency legislation in California, New York and Illinois in 2025, but that this alone is not enough at the current stage.

His proposed model is based on the U.S. Federal Aviation Administration approach. Frontier AI models trained using computing resources above a certain scale should be subject to mandatory third-party evaluations across four risk areas: cybersecurity, biological weapons, loss of AI control and the acceleration of automated research and development. He said the government should be able to block or cancel deployment depending on the evaluation results. Anthropic also released legislative proposals on frontier model testing and a policy framework to respond to job displacement.

Amodei said AI will bring more fundamental change to commerce than the internet or mobile. He added that regulation at a level applied to technologies such as cars, aircraft and pharmaceuticals, which can cause large-scale damage if poorly designed or operated, is appropriate for now. He said that if more powerful AI systems emerge, rules comparable to those for nuclear materials may be needed.

He compared the gap between AI's rapid advances and the pace of policymaking to Treebeard in "The Lord of the Rings." Just as hobbits try to awaken Treebeard to protect the forest but his response is too slow, AI evolves quickly while policy bodies move slowly, he wrote. Amodei said "Treebeard and his forest are waking up" and stressed that now is a window for action.