The former OpenAI VP of research explains why he walked away to build a competitor focused on AI safety
Dario Amodei sat down with Bloomberg to discuss a topic that has become one of Silicon Valley’s defining breakup stories: why he left OpenAI, the organization he helped build, to launch a direct competitor.
The split that reshaped the AI industry #
Amodei joined OpenAI in 2016, not long after the organization was founded. He eventually rose to VP of Research, placing him at the center of some of the most consequential decisions in early AI development.
By late 2020, he was done. His departure was officially announced on December 29, 2020, and he wasn’t alone. Roughly a dozen researchers, including his sister Daniela Amodei, walked out with him.
The core tension, as Amodei has framed it, came down to a gap between what OpenAI said it valued and what it actually did. He saw contradictions between the organization’s stated commitment to safety and alignment and the direction its leadership, particularly CEO Sam Altman, was steering the company.
In early 2021, Amodei and his team founded Anthropic with a thesis that you could build AI systems that were both highly capable and genuinely safe. The company has since developed the Claude family of models, which compete directly with OpenAI’s GPT products across consumer and enterprise markets.
Amodei has consistently cited two driving forces behind the split. First, a deep belief in scaling laws, the idea that making models bigger and training them on more data would yield dramatic improvements in capability. Second, the conviction that safety research needed to be baked into the development process from day one, not bolted on as an afterthought.
What this means for investors #
On the crypto side, the situation has gotten complicated. Tokenized pre-IPO instruments tied to Anthropic have circulated on the Solana blockchain, giving traders synthetic exposure to the company’s valuation. In May 2026, Anthropic warned that unauthorized transfers of these tokenized instruments would be declared void. The result was a 30% drop in the value of related tokens.
When the underlying company doesn’t recognize or authorize the instruments being traded, holders have essentially zero legal protection. It’s synthetic exposure with real downside and no guaranteed upside.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our