OpenAI Co-Founder Andrej Karpathy Joins Anthropic to Lead Claude Pretraining

Andrej Karpathy, one of the most recognized names in AI research, announced Monday that he is joining Anthropic’s pretraining team, where he will build a new group focused on using Claude itself to accelerate pretraining research, according to CNBC. Karpathy will work under team lead Nick Joseph on the large‑scale training runs that give Claude its core knowledge and capabilities.

“I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D,” Karpathy wrote in a post on X. He added that he remains “deeply passionate about education” and plans to return to that work “in time,” leaving the future of his AI education startup Eureka Labs uncertain.

The hire was immediately recognized as a major coup. “The hire is a major coup for Anthropic in the escalating competition for elite AI talent,” Axios wrote. Anthropic spokesperson told ³ that Karpathy starts this week.

Karpathy is one of the few AI researchers who can claim elite credentials across research, industry, and education. He was a founding member of OpenAI in 2015, where he worked on deep learning and computer vision research. In 2017, Elon Musk personally recruited him to lead Tesla's computer vision team for Autopilot — Musk described him in a court exhibit as “arguably the #2 guy in the world in computer vision” behind Ilya Sutskever, CNBC reported.

At Tesla from 2017 to 2022, Karpathy led the Autopilot AI team. After leaving, he briefly returned to OpenAI, then founded Eureka Labs, an AI education company. He also became one of AI's most influential public educators through his YouTube course “Neural Networks: Zero to Hero” and coined the term “vibe coding.” Recently he described himself as being in a “state of AI psychosis” since December, aggressively stress‑testing frontier models, according to Axios.

Karpathy holds a PhD in computer science from Stanford University and is one of the most‑followed AI researchers on social media, giving Anthropic not just technical talent but significant public visibility.

The most interesting detail in Karpathy's role isn't just that he's joining the pretraining team — it's what he's building. He will lead a team focused on using Claude to accelerate pretraining research, a form of AI‑assisted AI development. This is a bet on recursive self‑improvement: using today's models to build tomorrow's models faster.

Shelly Palmer described it as “a bet on recursive self‑improvement. Maybe we haven't hit the pre‑training ceiling yet. Maybe we just haven't thrown the right tools at it,” per his analysis.

The approach aligns with a broader industry trend toward AI researching AI. Google DeepMind has published extensively on using AI to improve chip design and training efficiency. OpenAI has hinted at automated alignment research. Anthropic is now putting one of the field's most celebrated researchers in charge of turning Claude into a tool for building better Claudes — a flywheel that could compound quickly if it works.

Karpathy isn't arriving alone. Anthropic has been on a hiring tear. Earlier in May, Ross Nordeen, a former Tesla engineer and founding member of Elon Musk's xAI, also joined Anthropic, CNBC reported. On the same day as Karpathy's announcement, cybersecurity veteran Chris Rohlf — with 20+ years of experience at Yahoo's elite “Paranoids” team and Meta — joined Anthropic's frontier red team to stress‑test advanced models against severe threats, TechCrunch reported.

"We have a real opportunity in front of us to dramatically improve cyber security with AI. I can't think of a better company or team to join at this critical moment in time," Rohlf wrote,.³

The pattern is clear: Anthropic is building depth across pretraining (Karpathy), infrastructure (Nordeen), and safety (Rohlf) — the three pillars of frontier AI development — just as both Anthropic and OpenAI race toward IPOs.

The Karpathy hire lands in the middle of an extraordinary competition for AI talent. With both Anthropic and OpenAI targeting IPOs in late 2026, every high‑profile hire carries outsized signaling value for future public‑market investors. Karpathy doesn't just bring technical expertise — his name on the S‑1 filing tells institutional investors that Anthropic can attract the field's most elite researchers away from OpenAI.

The competition extends beyond just these two. Google DeepMind recently hired Contextual AI researchers and their CEO in an $80 million to $90 million licensing deal, Silicon UK reported. Every major lab is racing to lock down the small pool of researchers capable of advancing the frontier — and the prices keep rising.

For builders, the talent concentration at a few labs cuts both ways. On one hand, it means frontier models keep getting better, faster. On the other, it means the gap between the top 3‑4 labs and everyone else is widening — and the APIs those labs control become ever more critical infrastructure.

Karpathy's specific focus — pretraining, not fine‑tuning or product — signals that Anthropic believes the next big leap in Claude's capabilities will come from better foundational training, not just better prompting or orchestration layers on top. Pretraining is "one of the most expensive, compute‑intensive phases of building a frontier model," TechCrunch noted.

Putting Karpathy — who has spent years obsessively stress‑testing frontier models and teaching neural network fundamentals — in charge of this phase suggests Anthropic sees untapped efficiency gains in how models learn their core knowledge. If his team succeeds, Claude models could improve in capability‑per‑dollar faster than competitors who focus on scaling compute alone.

For developers building on Claude's API: expect the model to get smarter at the foundational level, not just more feature‑rich at the application layer. The difference matters. A smarter base model makes every downstream use case better — coding, analysis, reasoning, and agentic behavior — without requiring developers to change anything.