Anthropic is founded

Anthropic was announced on 28 May 2021 with $124 million in funding and eleven founders, most of whom had left OpenAI at the end of 2020. Dario Amodei had led OpenAI's research, including GPT-2 and GPT-3, and Jared Kaplan had written the scaling laws; Chris Olah had led interpretability work at Google and OpenAI. The company described itself as an AI safety and research company and said it would build reliable, interpretable and steerable systems, and study frontier models to understand them.

Its founding thesis was that the same organisation should be at the frontier and be focused on safety, because safety research needed the most capable models to study, and because someone at the frontier had to be. Critics called this a contradiction; the company called it the only workable position. It structured itself as a public-benefit corporation with a trust intended to hold the founders to the mission.

Its first model, Claude, was released in March 2023, trained with a method the company called constitutional AI in which the model critiques its own outputs against written principles. By 2025 Anthropic was one of the three or four laboratories setting the frontier, and its founding is the origin of the safety vocabulary that governments adopted.