Dario Amodei believes that mechanistic interpretability, the science of looking inside AI models, is the most important technical breakthrough still needed to make frontier AI reliably safe and controllable.

Anthropic pioneered the science of mechanistic interpretability for looking inside AI models.

Dario Amodei joined OpenAI in mid-2026.

Dario Amodei has been watching the AI field for 15 years and working in it for 10 years.