OpenAI co-founder Andrej Karpathy joins Anthropic to lead pre-training research team

Andrej Karpathy, the prominent AI researcher who co-founded OpenAI and previously led Tesla’s Autopilot and Full Self-Driving programs, has joined Anthropic to work on pre-training research. Karpathy announced the move on X Tuesday, calling the next few years at the frontier of large language models especially formative.

A strategic hire for Anthropic

Karpathy started this week at Anthropic, where he is working under team lead Nick Joseph on pre-training — the computationally intensive phase responsible for giving Claude its core knowledge and capabilities. An Anthropic spokesperson confirmed to Bitcoin World that Karpathy will establish a new team focused on using Claude itself to accelerate pre-training research.

This hire signals Anthropic’s belief that AI-assisted research, rather than simply scaling compute, is the key to staying competitive with rivals like OpenAI and Google. Karpathy is one of the few researchers who bridges the gap between theoretical understanding of large language models and the practical realities of large-scale training runs.

Karpathy’s career arc

Karpathy’s journey through the AI industry has been closely watched. He left OpenAI in 2017 to join Tesla, where he led the company’s Full Self-Driving and Autopilot programs until 2022. He returned to OpenAI for a year before departing again in 2024 to launch Eureka Labs, a startup focused on applying AI assistants to education. Since then, he has shared few updates on Eureka Labs, and it remains unclear whether he will continue that venture alongside his new role at Anthropic.

He has also maintained an active presence in AI education through his online course Neural Networks: Zero to Hero and his YouTube channel, where he posts lectures on LLMs and AI. In his announcement, Karpathy said he remains deeply passionate about education and plans to resume that work in time.

Anthropic strengthens security team

Separately, Anthropic has brought on Chris Rohlf to its frontier red team, which stress-tests advanced AI models against severe threats. Rohlf, a cybersecurity veteran with over 20 years of experience, previously worked at Yahoo’s well-known security team known as The Paranoids and spent six years at Meta. He was also a fellow at Georgetown’s Center for Security and Emerging Technology, where he contributed to the CyberAI project.

In a post on X, Rohlf said there is a real opportunity to dramatically improve cybersecurity with AI and that he could not think of a better company or team to join at this critical moment.

What this means for the AI landscape

Karpathy’s move to Anthropic, combined with the addition of a seasoned cybersecurity expert, suggests the company is investing heavily in both frontier model development and safety research. Pre-training remains one of the most expensive and compute-intensive phases of building advanced AI systems, and Anthropic’s decision to focus on using AI to accelerate that work could give it a unique advantage.

For readers following the AI industry, this development underscores a broader trend: leading AI labs are increasingly competing not just on raw compute power, but on the quality of their research teams and their ability to innovate in how models are built and trained.

Conclusion

Andrej Karpathy’s return to frontier AI research at Anthropic, alongside the company’s parallel investment in cybersecurity expertise, reflects a dual focus on capability and safety. As the race to build more advanced language models intensifies, the composition of research teams and the methods they use to accelerate progress will likely become as important as the scale of the hardware they deploy.

FAQs

Q1: What will Andrej Karpathy do at Anthropic?
He will lead a new team focused on using Claude to accelerate pre-training research, working under team lead Nick Joseph. Pre-training is the phase that gives AI models their core knowledge and capabilities.

Q2: Why is this hire significant?
Karpathy is one of the few researchers with deep experience in both the theory and large-scale practice of training LLMs. His move signals Anthropic’s strategy of prioritizing AI-assisted research over simply scaling compute.

Q3: Will Karpathy continue his education work?
He has said he remains deeply passionate about education and plans to resume that work in time, but has not provided specific details. His startup Eureka Labs has not shared recent updates.