Andrej Karpathy defects from OpenAI, joins Anthropic's pre-training team
OpenAI co-founder moves to Anthropic to work on next-gen pre-training methods.
Andrej Karpathy, the renowned AI researcher and co-founder of OpenAI, announced on May 22, 2026, that he is joining Anthropic's pre‑training team. Karpathy, who previously led Tesla's Autopilot vision and later returned to OpenAI, will now focus on advancing pre‑training methods at Anthropic—the lab known for its safety‑focused Claude models. This defection is a significant coup for Anthropic, signaling their aggressive push to attract top talent as they race against OpenAI, Google, and Meta. The move also underscores a broader shift: top researchers are increasingly drawn to Anthropic's unique blend of cutting‑edge AI research and its commitment to alignment and safety.
Anthropic's recent momentum is clear. Co‑founder Jack Clark, speaking at Oxford, predicted an AI‑assisted Nobel Prize within 12 months while also acknowledging a non‑zero chance of catastrophic AI risk. Meanwhile, Google unveiled Gemini 3.5 Flash, a proactive agent called Spark, and a from‑scratch Search rebuild at I/O—backed by up to $190 billion in capital expenditures this year. Meta's Mark Zuckerberg was caught on leaked audio saying the company trained AI on its own employees before laying off 7,800 people. Elon Musk lost his lawsuit against Sam Altman on statute of limitations grounds. Against this backdrop, Karpathy's move highlights how talent decisions shape the AI landscape more than any single model release.
- Andrej Karpathy, OpenAI co-founder, joins Anthropic's pre-training team.
- The move signals Anthropic's aggressive push to lead in AI pre-training research.
- Karpathy's expertise could accelerate Anthropic's development of more capable and aligned models.
Why It Matters
Top AI talent moving to Anthropic highlights the escalating race for foundational AI research supremacy.