Anthropic CEO Dario Amodei calls for the AI industry to slow the development of frontier models, arguing that the pace is too fast relative to safety preparedness. At a conference in San Francisco, he uses a driving analogy about responding to safety incidents, saying companies should review their own practices rather than focus only on competitors. Several outlets also report Amodei warns that highly capable “agents” could quickly expand in scope, including potential risks to internet systems.

Outlets describe Amodei’s proposal as a structured approach to “pace” progress. The Verge says he offers a three-step plan and includes plans to bring in third-party evaluators such as giving access to METR to support assessment of safety practices and commitments. TechRadar and others say his concerns extend to how quickly agent capabilities could scale. However, Bloomberg’s opinion coverage argues that Amodei’s proposal is “quite vague” on how slowing would work in practice, while other reports note that the idea has generated disagreement.

Across coverage, Nvidia CEO Jensen Huang is reported to argue against deceleration, and NBC notes the broader debate involves top leaders including those from OpenAI and other prominent figures. NDTV and other summaries describe wider discussion about extinction-level and existential risks, alongside claims of partial alignment among some leaders even as implementation details remain contested.