Anthropic is founded by departing OpenAI researchers

Anthropic announced its first funding round — $124 million in Series A financing led by Jaan Tallinn, with participation from Dustin Moskovitz and Eric Schmidt among others — the announcement that introduced the company publicly. The company itself had been incorporated some months earlier, in early 2021, by siblings Dario and Daniela Amodei together with several colleagues who had left OpenAI around the same time; Dario Amodei had been OpenAI’s vice president of research and Daniela its vice president of safety and policy. Other co-founders, including Jared Kaplan, Chris Olah, Tom Brown, Sam McCandlish, Jack Clark and Ben Mann, had worked on GPT-3, scaling laws and safety research there.

The founders described their departure as motivated by disagreement over how OpenAI was weighing safety against the pace of capability development. Anthropic positioned itself from the outset as a research company built around that disagreement, stating its work would focus on making large-scale AI systems more reliable, interpretable and steerable, with safety research treated as core to the roadmap rather than a separate function.

The announcement gave few specifics about products — Anthropic released no public model for another year and a half — but its personnel signalled where its research would concentrate: interpretability, drawing on Olah’s earlier work on visualising neural networks, and techniques for training models to follow instructions and avoid harmful outputs, which later surfaced as Constitutional AI and the company’s Responsible Scaling Policy. Anthropic’s founding marked the first significant split of frontier AI talent explicitly over safety disagreements, a pattern of departures and spinouts — from OpenAI and later from Anthropic itself — that recurred repeatedly across the industry in the years that followed.