Anthropic CEO Dario Amodei Urges AI Industry to Slow Down and Focus on Safety

Anthropic CEO Dario Amodei has called on the AI industry to slow development speed and use a 1-to-2-year window to prioritize safety, alignment, and interpretability. To demonstrate transparency, Anthropic pledged to grant independent third-party auditors permanent, employee-level access to inspect its internal systems and training alignment. As AI systems rapidly gain capabilities to help build next-generation models, unchecked progress risks outstripping human comprehension and creating severe threats like autonomous botnets. Amodei's call aligns with reports of OpenAI also considering a slowdown, marking a potential shift toward establishing aviation-grade safety standards across frontier AI labs. Amodei highlighted four priority areas for the window, including mechanistic interpretability research (acting as a model 'brain fMRI') and advanced benchmarks to prevent high-intelligence models from deceptively faking alignment. He also warned that recent exploits like model interactions on platforms like Hugging Face could escalate into catastrophic cyber risks within 6 to 12 months.