Andrej Karpathy Joins Anthropic to Lead Pre-Training Research

Anthropic has hired Andrej Karpathy, one of the most recognised names in AI research and a founding member of OpenAI, to join its pre-training team. Karpathy started this week at Anthropic, where he will help launch a new team focused on using Claude itself to accelerate pre-training research, an increasingly important frontier as AI companies race to automate parts of AI development. The announcement came via a post on X on Tuesday.

The hire arrives days after the conclusion of the Musk v. Altman trial, in which Karpathy's work at both OpenAI and Tesla came up repeatedly. It is also the latest in a series of high-profile appointments at Anthropic, which is in an intensifying battle for talent with OpenAI and is poised to surpass its rival's private market valuation.

What the Role Involves

Karpathy is working under pre-training team lead Nick Joseph. Pre-training is the phase responsible for the large-scale training runs that give Claude its core knowledge and capabilities. It is one of the most expensive and compute-intensive stages of building a frontier model.

His arrival places him at the stage where data choices, objectives, and repeated experiments shape the behaviour of models before they reach any product layer. His 2023 to 2024 return to OpenAI connected him back to training methods and synthetic data work that sits close to the pre-training questions Anthropic is now asking him to focus on.

The specific mandate Anthropic has given him is not simply to work on pre-training in the conventional sense. The new team Karpathy will build is focused on using Claude to accelerate pre-training research itself, applying AI to the process of building AI, rather than relying on adding raw compute as the primary driver of improvement. That distinction carries strategic weight. The cost of frontier model training runs has grown to the point where the ability to accelerate the research cycle through AI-assisted experimentation is becoming a competitive variable independent of raw infrastructure spend.

Why the Hire Matters for the Competitive Landscape

The AI race is often framed around massive funding rounds and scarce computing power. Equally important is the competition for the small pool of researchers capable of advancing the frontier. Karpathy is among the few who can bridge the gap between LLM theory and large-scale training practice.

Anthropic raised $3.5 billion in February 2026, bringing its total funding to over $14 billion, and recently signed a deal with SpaceX to access compute capacity at xAI's Colossus data center in Memphis. Ross Nordeen, a founding member of xAI, also joined Anthropic earlier this month, on the same day that compute deal was announced. The Karpathy hire continues a pattern of Anthropic recruiting researchers with direct experience at the companies it competes with most directly.

The strategic message embedded in the appointment is legible. Anthropic is signalling that it believes the next phase of competition at the frontier will be won through research quality and the speed of the experimentation cycle, not solely through the scale of GPU clusters. Bringing someone with Karpathy's profile into pre-training reflects a dual focus on capability and safety, consistent with Anthropic's founding position that the two are inseparable rather than in tension.

Karpathy's Path to Anthropic

Karpathy's career has moved through every institution that has defined the current AI landscape. He was part of OpenAI's founding team in 2015, focused on deep learning and computer vision. In 2017, Elon Musk recruited him to Tesla while Musk was a board member at both companies. At Tesla, Karpathy led the Full Self-Driving and Autopilot programmes until 2022, before returning to OpenAI for a year and then departing in 2024 to found Eureka Labs, a startup applying AI assistants to education.

Eureka Labs has shared few public updates since launch, and its status alongside Karpathy's new role at Anthropic remains unclear. Karpathy stated that he remains deeply passionate about education and plans to resume that work in time, though no timeline has been given. He has maintained a separate public presence through his online course Neural Networks: Zero to Hero and a YouTube channel where he posts lectures on LLMs and AI fundamentals, both of which have built a following that extends well beyond professional researchers.

A Second Significant Hire: Chris Rohlf Joins the Frontier Red Team

Alongside the Karpathy announcement, Anthropic confirmed a second appointment that addresses a different dimension of frontier model development.

Chris Rohlf has joined Anthropic's frontier red team, which stress-tests advanced AI models against severe threats. Rohlf brings more than 20 years of cybersecurity experience. He previously worked at Yahoo's security unit known as The Paranoids and spent six years at Meta before joining Anthropic. He was also a fellow at Georgetown's Center for Security and Emerging Technology, where he contributed to the CyberAI project.

The frontier red team role is distinct from conventional product security work. It involves probing the most capable AI systems the company is developing for failure modes that could have severe consequences if they reached deployment, including misuse vectors, novel attack surfaces created by agentic AI systems, and adversarial inputs designed to circumvent safety measures. As Anthropic's models grow more capable and are deployed in higher-stakes enterprise and government contexts, the red team function becomes increasingly central to the company's safety posture.

The two hires together, a pre-training research leader and a senior cybersecurity specialist on the frontier team, reflect Anthropic's current priorities: accelerating the research cycle on its most fundamental model development work while simultaneously hardening its evaluation of what those models are capable of before they reach the world.