Andrej Karpathy Joins Anthropic to Lead Claude Pre-Training Research
Summary: Andrej Karpathy announced on May 19, 2026 that he is joining Anthropic, where he will lead a new pre-training research team that uses Claude itself to design the experiments and infrastructure behind future Claude models.
Key facts
- Announcement: Karpathy posted on X on May 19, writing that "the next few years at the frontier of LLMs will be especially formative" as his reason for joining
- Role: Reporting to Nick Joseph (Anthropic's head of model development), leading a newly formed team focused specifically on using Claude to accelerate pre-training research
- Recursive mandate: The team will apply current best Claude models to hypothesis generation, experiment design, literature review, and evaluation infrastructure — AI training the next AI
- Background: OpenAI founding member → Tesla FSD/Autopilot lead (2017–2022) → OpenAI (one year) → Eureka Labs (education startup, now paused)
Why it matters
Karpathy is one of the few people in deep learning who has operated at the frontier of research, production systems, and public education simultaneously. His hire signals that Anthropic is treating pre-training capability as a top strategic priority as it heads toward an IPO — and it represents a symbolic talent loss for OpenAI. The race for top AI researchers is now as consequential as the race for compute.
Read more
How this story unfolded
- Claude Fable 5 Lands in Microsoft 365 Copilot as Multi-Model Enterprise AI Arrives
- U.S. Slaps Export Controls on Claude Fable 5 After Safety Guardrail Bypass
- Claude Fable 5 Goes Public — Then U.S. Export Controls Suspend Access Three Days Later
- Anthropic: Claude Now Writes 80% of Its Own Code — Company Calls for Global AI Pause