OpenAI co-founder Andrej Karpathy joins Anthropic’s pre-training team

Details

Andrej Karpathy announced on X that he has joined Anthropic, starting this week on the pre-training team under lead Nick Joseph. Pre-training is the compute-intensive phase responsible for giving Claude its core knowledge and capabilities. According to an Anthropic spokesperson, Karpathy will start a dedicated team focused on using Claude to accelerate pre-training research itself — effectively an AI-assisted research group within the pre-training division. Karpathy's background spans: co-founding OpenAI and doing early deep learning research there (until 2017), leading Tesla's Autopilot and Full Self-Driving programs (2017–2022), returning to OpenAI for one year, then leaving in 2024 to start Eureka Labs — an AI-in-education startup. Karpathy has indicated he remains committed to education and plans to resume that work eventually, though the current status of Eureka Labs is unclear. The fact that Karpathy chose Anthropic over a return to OpenAI is a significant talent signal. His stated reason — that 'the next few years at the frontier of LLMs will be especially formative' — aligns with Anthropic's narrative of building responsibly at the frontier. The move is also strategically notable: using Claude to accelerate pre-training research suggests Anthropic believes AI-assisted R&D is a competitive lever, not just compute scale. Separately, Anthropic also hired Chris Rohlf to its frontier red team — a 20-year cybersecurity veteran previously from Yahoo ('The Paranoids') and Meta, and a Georgetown fellow on the CyberAI project.