Andrej Karpathy joined Anthropic this week to lead a new team building AI-accelerated pretraining research tools, the company confirmed Tuesday. The OpenAI co-founder and former Tesla AI chief will report to pretraining lead Nick Joseph and focus on using Claude to speed up the large-scale training runs that give the model its core capabilities. Pretraining remains the most compute-intensive and expensive phase of frontier model development.
“I've joined Anthropic. 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.”— Andrej Karpathy, AI researcher and OpenAI co-founder
Karpathy announced the move on X, saying the next few years at the frontier of large language models will be especially formative. He started at Anthropic on May 19, 2026, after leaving his education-focused startup Eureka Labs. The hire is a major win for Anthropic in the high-stakes competition for elite AI talent.
The researcher is one of the few who can bridge LLM theory and large-scale training practice. At OpenAI, Karpathy focused on deep learning and computer vision until he left in 2017 to join Tesla. He led Tesla's Full Self-Driving and Autopilot programs before departing in 2022, then returned to OpenAI for 1 year before leaving again in 2024 to start Eureka Labs.
Key facts
- 01Andrej Karpathy joined Anthropic this week to lead a team using Claude to accelerate pretraining research.
- 02Karpathy co-founded OpenAI, led Tesla's Full Self-Driving and Autopilot programs, and returned to OpenAI for 1 year before leaving in 2024.
- 03Chris Rohlf joined Anthropic's frontier red team with 20 years of cybersecurity experience, including 6 years at Meta.
- 04Karpathy will work under Nick Joseph on pretraining, the most compute-intensive phase of building frontier models.
- 05The hire signals Anthropic's bet on AI-assisted research over pure compute to compete with OpenAI and Google.
Anthropic's decision to tap Karpathy for AI-assisted research rather than pure compute expansion is a clear signal about how the company plans to stay competitive with OpenAI and Google. The move suggests Anthropic believes research velocity, not just raw compute, will decide the next generation of frontier models. Karpathy has not shared many updates on Eureka Labs since its 2024 launch, and it remains unclear whether he will continue with the startup.
He has also taught an online course called Neural Networks: Zero to Hero and maintains a YouTube channel with lectures on LLMs and AI. His educational work has made him one of the most recognizable figures in AI research outside the major labs.
Separately, Anthropic brought on Chris Rohlf to its frontier red team, which stress-tests advanced AI models against severe threats. Rohlf joins with 20 years of cybersecurity experience, including 6 years at Meta and prior work at Yahoo's cybersecurity team known as The Paranoids. He was also a fellow at Georgetown's Center for Security and Emerging Technology, where he worked on the CyberAI project.
“I remain deeply passionate about education and plan to resume my work on it in time.”— Andrej Karpathy, AI researcher and OpenAI co-founder
The frontier red team hire underscores Anthropic's continued investment in safety and security as models grow more capable. Rohlf's background in offensive security and AI-driven cybersecurity research positions him to probe Claude for vulnerabilities at scale.
Anthropic has been aggressive in recruiting top-tier technical talent as it scales Claude and competes for enterprise contracts. The company recently acquired Stainless for over $300M, pulling the SDK tool from OpenAI and Google in a move that signaled its intent to own more of the developer experience. Karpathy's hire fits the same pattern — Anthropic is betting that the researchers who built the first generation of frontier models will be critical to building the next.
The risk for Anthropic is execution. Karpathy has a track record of departures — twice from OpenAI, once from Tesla — and his ability to scale a team inside Anthropic's structure remains untested. But the upside is clear: if Claude can accelerate its own pretraining research, Anthropic gains a compounding advantage over labs still relying on human-paced iteration.
For Karpathy, the move back into frontier research suggests he sees the next few years as a critical inflection point. His work at Anthropic will likely focus on closing the loop between research and deployment at the scale where even small efficiency gains translate to tens of millions in compute savings. Whether that bet pays off will depend on how quickly his team can turn Claude into a tool that makes itself better.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
