Dario Amodei

Some people walk away from the safe path because they see something others haven't noticed yet. Dario Amodei was vice president of research at OpenAI, one of the most influential roles in artificial intelligence, when he decided to leave. Not for money, not over an ego dispute. He left because he was convinced that building increasingly capable AI systems without placing safety at the center of every decision was a dangerously reckless bet. In 2021, alongside his sister Daniela and a group of OpenAI colleagues, he co-founded Anthropic.

At Anthropic, Amodei oversaw the development of the Claude family of models, designed from the ground up around a philosophy called "Constitutional AI": instead of training the AI solely on direct human feedback, you give it a set of principles and let it evaluate its own responses. It's a technical idea, yes, but also a statement of values: that an AI system can learn to reason about right and wrong without relying entirely on an external judge to flag every mistake. Anthropic's foundational paper on this technique, published in 2022, became a required reference across the field.

The impact came quickly. Time named him among the 100 most influential people in the world in 2025, and again in 2026 alongside Daniela. Anthropic reached an estimated valuation of $965 billion, positioning itself as one of the most important AI companies on the planet. But perhaps the most revealing thing about Amodei is what he writes: his long essays on the risks and possibilities of AI, like "Machines of Loving Grace" published in 2024, show someone who isn't selling a product but trying to think out loud about one of the hardest problems of our time.

What I find admirable about Dario is that he chose discomfort. He could have stayed where he was, kept building larger and larger models, and taken the credit. Instead, he bet on the hard question: not just "can we do this?" but "should we, and how?" Every time you interact with Claude, there's a little piece of that bet in every response, in every guardrail, in every time the model says "I don't know" instead of making something up.