Andrej Karpathy, one of the most influential figures in modern AI, announced on May 19, 2026 that he has joined Anthropic. The OpenAI co-founder and former director of AI at Tesla will work on pre-training under team lead Nick Joseph and start a new group focused on using Claude itself to accelerate the next generation of large-scale training runs.
Key Highlights
- Karpathy is joining Anthropic's pre-training team, working under Nick Joseph
- He will lead a new effort to use Claude to accelerate pre-training research
- Anthropic also brought on cybersecurity veteran Chris Rohlf to its frontier red team
- The move signals Anthropic's bet on AI-assisted research as a competitive edge
- Karpathy says he remains passionate about education and plans to resume that work later
A Quiet Tweet, a Loud Signal
"Personal update: I've joined Anthropic," Karpathy wrote in a short post on X that quickly became one of the most discussed AI announcements of the year. "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."
The brevity is classic Karpathy, but the implications are anything but small. He co-founded OpenAI in 2015, returned for a second stint, then left in 2024 to launch Eureka Labs, an education startup focused on AI tutors. His decision to step back into frontier research, and to do it at Anthropic rather than OpenAI, is being read across the industry as a meaningful talent-market signal.
What the New Team Will Do
According to an Anthropic spokesperson cited by TechCrunch, Karpathy will "start a team focused on using Claude to accelerate pre-training research." Pre-training is the foundational, compute-heavy phase that gives a model like Claude its core knowledge and capabilities. By turning the model on itself, Anthropic is doubling down on the idea that AI-assisted research, not just raw compute, is how labs stay competitive with OpenAI and Google.
That thesis fits Anthropic's broader product trajectory. The company has been pushing Claude as a research and engineering partner through tools such as Claude Code, managed agents, and increasingly autonomous coding workflows.
Why It Matters
Karpathy is rare in AI: a researcher who is also a generational educator. His Stanford lectures on neural networks and his recent series rebuilding GPT from scratch shaped how an entire cohort of engineers understands modern deep learning. Putting that pedagogical clarity on the inside of a frontier pre-training team gives Anthropic both technical depth and cultural pull.
The hire also lands during a period of unusual movement at the top of the field. Anthropic has been on a sustained hiring run, including security and safety leadership, and recently closed a 30 billion dollar Series G at a 380 billion dollar valuation. Adding Karpathy reinforces the narrative that the company is positioning itself as the primary research-led alternative to OpenAI.
Background
Karpathy's career has tracked the modern AI era almost exactly. After a PhD at Stanford under Fei-Fei Li, he joined OpenAI as a founding member in 2015. In 2017 he left for Tesla, where he led the Full Self-Driving and Autopilot vision teams for five years. He returned briefly to OpenAI in 2023 before leaving in 2024 to found Eureka Labs.
In between, he has become one of the most-followed independent voices in AI on X, with widely circulated essays on vibe coding, agentic workflows, and the future of software engineering.
What's Next
Karpathy says education work will resume "in time," suggesting Eureka Labs is not being wound down. For Anthropic, the immediate question is what Claude-accelerated pre-training actually looks like in practice, and whether it can compress the cost and timeline of the next Claude generation. Expect the first concrete signals to appear in benchmark scores and model release cadence over the coming year.
Source: TechCrunch