Karpathy joins Anthropic’s pretraining team in major frontier talent move

TL;DR:

  • Andrej Karpathy — OpenAI founding member, former Tesla autopilot and AI lead, and one of the most-followed AI researchers online — has joined Anthropic’s pretraining team this week.
  • He sits under pretraining head Nick Joseph and will work on the large-scale training runs that produce Claude’s core capabilities, while continuing his Eureka Labs education work separately.
  • The hire follows John Schulman’s 2024 move from OpenAI to Anthropic and is the latest senior departure from the ChatGPT maker, after Ilya Sutskever and Mira Murati.

Karpathy confirmed the move on X on Tuesday, saying he was “very excited to join the team here and get back to R&D” and that he expected the next few years at the frontier of large language models to be “especially formative”. Anthropic placed him on the pretraining team — the part of the organisation that runs the large-scale training jobs whose architecture and data choices determine what each generation of Claude can do.

A clear pattern of OpenAI departures

The hire continues a steady flow of senior OpenAI alumni into Anthropic and adjacent firms. John Schulman, another OpenAI co-founder, left for Anthropic in 2024. Former chief scientist Ilya Sutskever now runs Safe Superintelligence; former CTO Mira Murati founded Thinking Machines. Karpathy himself studied under Stanford’s Fei-Fei Li before joining OpenAI as an early staff member and then leading Tesla’s autopilot and AI work until 2022, when he stepped away to focus on independent research and education. He announced Eureka Labs, an AI-integrated education platform, in 2024; he has said this remains a long-term commitment alongside his new R&D role.

For Anthropic the strategic signal is straightforward: pretraining hires are infrastructure hires. Claude’s competitive position rests on training runs that are now in the same compute tier as GPT-5 and Gemini 3, and the small group of researchers who can architect those runs is the binding constraint on every frontier lab’s roadmap. Adding Karpathy alongside Joseph’s existing team sharpens that capacity at exactly the moment Anthropic is expanding into UK and European compute (its 800-seat London office was confirmed last month).

Looking forward

For UK readers, the hire matters because it deepens the talent gravity around Anthropic’s London expansion at a time when Inflection AI is re-establishing a UK entity and Google DeepMind continues to anchor the Kings Cross cluster. Karpathy’s R&D output has historically been highly public — his lectures, GitHub repositories and “from scratch” educational videos have shaped a generation of practitioners — so UK researchers and students should expect continued open output even as he now does proprietary work at Anthropic. The competitive picture for OpenAI is harder. The company has lost three of its most-cited research figures in eighteen months.

Share this article