Andrej Karpathy has joined Anthropic’s pre-training team, a move that puts one of AI’s most recognizable researchers back inside the frontier model race.
The hire matters because Karpathy is not just another senior name being shuffled between labs. He helped shape OpenAI’s early research culture, later led Tesla’s Autopilot and computer vision work, and has spent the past few years translating deep learning into plain language for everyone else.
According to TechCrunch, Karpathy started this week at Anthropic and will work on pre-training under team lead Nick Joseph. That is the part of model development where companies run the large-scale training jobs that give systems like Claude much of their core knowledge and capability, and it is also one of the most expensive phases in the business.
Karpathy confirmed the move himself on X, writing that the next few years at the frontier of large language models will be especially formative. He said he was excited to get back to research and added that he remains deeply committed to education, which suggests this is not a clean break from the teaching work that made him such an influential public figure in the first place.
The timing tells its own story. Anthropic has been pushing hard to narrow the gap with OpenAI and Google in the race to build stronger foundation models, and hiring a researcher with Karpathy’s background is a statement about how seriously it views that competition. TechCrunch reported that Anthropic plans to have him start a team focused on using Claude to accelerate pre-training research, which points to a broader bet on AI-assisted research rather than brute-force compute alone.
Karpathy has a rare profile in modern AI. He is a researcher, an engineer, a teacher and, in many ways, a public interpreter of the field’s most technical shifts. That mix is valuable at a moment when frontier labs are not only trying to build better models, but also trying to build better ways of building models.
At OpenAI, he was part of the founding group and worked on deep learning and computer vision before leaving in 2017. He then joined Tesla, where he led the company’s Full Self-Driving and Autopilot programs until 2022. After another return to OpenAI, he left again in 2024 to launch Eureka Labs, an AI education startup focused on applying assistants to learning.
That background makes him unusually relevant to pre-training, which is often discussed as if it were just a compute problem. In reality, it is a research problem wrapped around massive infrastructure, and small gains can matter a lot when they are multiplied across enormous training runs. A researcher who understands both the theory and the practical constraints can be useful in a way that is hard to replace.
It also helps explain why Anthropic would want him now, not later. The company has built a reputation for strong technical talent and safety work, but the race has become less about brand identity and more about who can keep improving faster. Bringing in a figure as visible and technically credible as Karpathy signals that Anthropic wants to keep widening its research bench, not merely keep pace.
The broader talent fight
Anthropic has been adding to its technical ranks as competition across the sector intensifies. TechCrunch also noted that the company recently brought on Chris Rohlf to its frontier red team, which stress-tests advanced models against severe threats. That kind of hiring suggests Anthropic is reinforcing both capability and safety at the same time, a combination that has become central to how frontier labs present themselves.
For OpenAI, the move is awkward in symbolic terms, even if no single hire decides the market. Karpathy is one of the best-known names in the company’s early history, and his return to a direct research role at a rival underscores how fluid the top end of AI talent has become. People who helped create the field are now circulating among its most powerful institutions, and each move is watched like a market signal.
For Anthropic, the upside is obvious. Elite researchers do not just write papers or ship features, they shape the direction of an entire lab. If Karpathy helps Anthropic use Claude to improve pre-training itself, that could create compounding advantages in model development, especially at a time when every leading company is trying to make its training pipeline more efficient and more adaptive.
He also brings audience reach. Karpathy has spent years making machine learning legible to outsiders through courses, lectures and long-form commentary. That public credibility matters in a field where recruiting, branding and technical authority are increasingly intertwined. When a researcher like that joins a company, it does more than fill a seat, it changes how the company is perceived by the people who might work there next.
He has not said much about the future of Eureka Labs, and TechCrunch reported that it is unclear whether he will continue with the startup. For now, the signal is simple: one of the defining engineers of the deep learning era is back in the lab where the next generation of foundation models is being built. In a market this crowded, that kind of move still carries weight.
Also read: South Korea’s push for mandatory AI watermarks forces startups to rethink product design • Demis Hassabis’s Anthropic stake complicates the AI power map • Meta shifts 7,000 workers into AI as it tightens its grip on the future
