Dario Amodei is the CEO and co-founder of Anthropic, the AI safety company. Previously VP of Research at OpenAI. He writes publicly on AI safety, frontier model development, and the institutional structures needed to govern AI capability growth.
Operating themes
- Pacing through verification, not halting Amodei argues that capability growth can continue responsibly only when safety assessments are made by third parties, not self-reported by the labs doing the training.
- Embedded evaluators as the credibility mechanism His proposal for embedded third-party evaluators resolves the conflict of interest in self-certification of safety commitments.
- Recursive self-improvement as the accelerant He identifies models assisting with training of the next generation as the development that changed his position on the urgency of pacing.
Cards in this library
- Third-party evaluators embedded in AI development are the only mechanism that makes pacing commitments verifiable: Third-party evaluators embedded in AI development are the only mechanism that makes pacing commitments verifiable rather than self-reported.