Andrej Karpathy, the prominent AI researcher who co-founded OpenAI and formerly led AI at Tesla, has joined Anthropic. He will be working under pre-training team lead Nick Joseph.
“I’ve joined Anthropic,” Karpathy posted on X. “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.” He added that he remains deeply passionate about education and plans to resume his work on it in time.
Pre-training is responsible for the large-scale training runs that give Claude its core knowledge and capabilities. It represents one of the most expensive and compute-intensive phases of frontier model development. An Anthropic spokesperson confirmed that Karpathy will head a new team focused on using Claude itself to accelerate pre-training research. This hiring signals Anthropic’s belief that AI-assisted research, rather than brute-force compute alone, is the key to competing with OpenAI and Google.
Karpathy is among a select group of researchers capable of bridging the gap between LLM theory and large-scale training practice. He previously focused on deep learning at OpenAI until leaving in 2017 for Tesla, where he led the Full Self-Driving (FSD) and Autopilot programs. He returned to OpenAI in 2022 for a year before departing in 2024 to launch Eureka Labs, an AI-education startup. It remains unclear whether he will continue running the startup alongside his new role. Karpathy is also famous for his online course "Neural Networks: Zero to Hero" and his educational YouTube channel.
Separately, Anthropic has recruited cybersecurity veteran Chris Rohlf to its frontier red team to stress-test advanced models against severe threats. Rohlf brings over 20 years of experience, having previously worked at Yahoo’s elite security team "The Paranoids," and most recently spending six years at Meta. He was also a fellow at Georgetown's Center for Security and Emerging Technology's CyberAI project.
[AgentUpdate Depth Analysis] Karpathy’s appointment to lead a team dedicated to using Claude to accelerate pre-training marks a pivotal transition towards "Agent-in-the-loop" model development. Traditionally, scaling frontier models relied on massive compute and human engineering. By leveraging Claude as an autonomous research Agent to optimize training configurations, clean data, and discover architectures, Anthropic is pioneering a self-improving feedback loop. This vertical integration of Agent capabilities into the core training pipeline could drastically reduce R&D costs and outpace brute-force scaling. Compared to competitors like OpenAI, Anthropic's emphasis on AI-assisted research indicates that the future of AI Agents lies not just in executing tasks for end-users, but in fundamentally driving the evolution and training of the next generation of foundation models.
