Andrej Karpathy joins Anthropic, returns to AI research and pre-training

Andrej Karpathy’s move to Anthropic puts one of the best-known names in modern AI back into the center of frontier model work, this time on pre-training, the phase that helps determine what large language models can do before they are tuned for specific tasks. The shift is more than a personnel change. It is another sign that the contest among leading AI labs is now as much about elite technical talent as it is about products, computing power and model scale.

Karpathy said he was joining Anthropic to get back to research and development, and said the next few years at the frontier of large language models would be especially formative. He also said he remains deeply passionate about education and plans to return to that work in time. The move came after a period in which he had been building Eureka Labs, an AI-native education startup.

His résumé gives the hiring news unusual weight. Karpathy was listed among OpenAI’s founding members in the company’s original 2015 announcement, alongside Trevor Blackwell, Vicki Cheung, Durk Kingma, John Schulman, Pamela Vagata and Wojciech Zaremba. Tesla later identified him as senior director of AI in materials for its 2019 Autonomy Investor Day, when Elon Musk and other executives were set to speak about the company’s autonomy push.

Anthropic, based in San Francisco, describes itself as an AI safety and research company focused on building reliable, interpretable and steerable systems. It also says it is a public benefit corporation dedicated to securing AI’s benefits and mitigating its risks. The company has become one of OpenAI’s most serious rivals and a magnet for respected technical talent as it scales its research organization and expands the Claude business.

That combination makes Karpathy’s arrival strategically significant. Pre-training remains one of the most consequential layers of frontier AI development because it shapes the core capabilities of the underlying model before downstream products are built on top of it. Bringing in a researcher who helped found OpenAI, led AI work at Tesla and then stepped into AI education suggests Anthropic is betting that the next phase of competition will be decided by people who can influence the foundations of the models themselves.