Andrej Karpathy Joins Anthropic to Accelerate AI Development

Andrej Karpathy, a founding member of OpenAI and former AI director at Tesla, announced on Tuesday, May 19, 2026, that he is joining Anthropic's pretraining team. Andrej Karpathy's high-profile recruitment immediately reconfigures the landscape of frontier AI development. His background includes foundational contributions to large language models.

A key figure from OpenAI's founding team now directly bolsters its fiercest competitor. The move of a key figure from OpenAI's founding team to its fiercest competitor intensifies the battle for foundational AI talent.

Anthropic is poised to significantly accelerate its core large language model development, potentially narrowing the gap with or even surpassing competitors in specific areas of model pretraining.

Karpathy's Role: Deep in Anthropic's Core AI

Andrej Karpathy, a founding member of OpenAI and former AI director at Tesla, has joined Anthropic's pretraining team, according to Business Insider. He will work under team lead Nick Joseph on the initial training of large AI models, The Decoder and TechCrunch reported. Karpathy's direct placement on the pretraining team underscores Anthropic's strategic intent: to bolster its foundational model capabilities with top-tier expertise. The focus on initial training implies a direct push for core model improvements, leveraging Karpathy's extensive prior experience in large language model development.

Accelerating Claude: A Strategic Mandate

Karpathy will lead a new team focused on using Claude to accelerate pretraining research, according to Fortune and The Decoder. While Bloomberg broadly reported Karpathy would 'focus on research and development,' his specific mandate involves optimizing existing models. Karpathy's focused effort targets accelerating pretraining, marking a strategic pivot towards optimizing existing large language models. Karpathy's specific mandate positions Anthropic to leverage Karpathy for direct innovation on its core Claude models, aiming for a distinct competitive edge in development efficiency over sheer model size.