How Dario Amodei's Departure from OpenAI Led to the Rise of Anthropic's Claude

Dario Amodei, a key researcher at OpenAI, left the company in 2020 due to differences in vision around AI safety and responsible development. He co-founded Anthropic, which developed a new training method called Constitutional AI that has made their model, Claude, outperform ChatGPT in coding tasks.

Details

Dario Amodei was a key research leader at OpenAI, helping to scale the early GPT models. However, by late 2020, he and a group of colleagues felt the company's direction no longer aligned with their own vision around AI safety and responsible development. They wanted to build interpretability, steerability, and strong alignment into the model from the very beginning, rather than adding fixes later. Leaving OpenAI, Amodei co-founded Anthropic in 2021 with the goal of creating the most capable AI while making it as safe and trustworthy as possible. Anthropic's secret is a training method called Constitutional AI, where the model is given a written 'constitution' of guiding principles to critique and improve its own outputs. This approach leads to fewer made-up facts, better reasoning, and more thoughtful and reliable outputs. By 2026, Anthropic has released an updated, more detailed version of Claude's constitution, further improving its ability to follow instructions precisely and push back helpfully when needed. This translates directly into Claude outperforming ChatGPT on coding benchmarks and developer polls, with its ability to produce cleaner, more robust code that is production-ready.