Anthropic CEO: Regulate AI Like Aircraft for Safety

Anthropic CEO Dario Amodei has written a new article titled "Policy on the AI Exponential." He uses "The Lord of the Rings" as a metaphor, comparing slow policymaking to the Ents and rapid AI evolution to an urgent threat. Amodei believes that if AI scaling laws continue, we could see "powerful AI" in one to two years. He notes that AI now writes most of the code for major AI companies, a significant change from four years ago. He proposes regulating AI like aircraft, with mandatory third-party testing for cutting-edge models. Specific solutions include audits for models exceeding a certain computing power in areas like network security and biological weapons. The government would have the right to prevent deployment of models with unacceptable risks. Amodei also suggests AI companies need strong security standards and must cooperate with the government. He warns that if future AI systems are like weaponizable nuclear materials, regulation will need to be even stronger. Economically, Amodei argues AI could bring unprecedented growth but also a significant labor shock. He suggests policies like wage insurance, tax incentives, and labor training subsidies. The bottom line is that Amodei is highlighting a critical gap between fast-evolving AI and slow-moving policy, urging action now to shape its future.