John Jumper, the scientist who led the team that built AlphaFold, is leaving Google DeepMind for Anthropic. He announced the move on X on June 19, saying he plans to “take some time to recharge” before starting at the frontier AI lab. Jumper spent nearly nine years at DeepMind, joining six months after finishing his PhD.
The hire is a statement. Anthropic is not just competing on language models. It is building a team capable of tackling hard scientific problems, the kind that require deep domain expertise and years of sustained work. Jumper is the public face of one of the most celebrated AI-for-science projects ever. AlphaFold solved a 50-year problem in biology: predicting protein structures from amino acid sequences. The system now covers over 200 million proteins, and the database is used by more than 2 million researchers worldwide.
Jumper shared the 2024 Nobel Prize in Chemistry for the work, alongside Demis Hassabis and David Baker. He was the youngest of the three laureates.
The timing matters. DeepMind merged with Google’s Brain team in 2023 to form Google DeepMind, a consolidation that brought together the company’s two major AI research units under Hassabis. That merger created a single large organization with overlapping priorities. Jumper’s departure suggests that the gravitational pull of a single lab, even one as well-resourced as Google DeepMind, is not enough to retain top-tier talent when frontier AI labs are competing on research culture, autonomy, and mission.
Anthropic has been building its science capabilities quietly. The company hired Tom Brown, a co-author of the original GPT-3 paper, in 2023. It brought on Jan Leike, formerly of OpenAI’s alignment team, in 2024. Jumper is a different order of hire. He brings a track record of leading a multi-year, high-stakes research project that delivered a real product. That is a rare combination in AI research, where many high-profile papers never leave the arXiv.
What Jumper will do at Anthropic is not yet public. The company has not announced a formal biology or science research division. But the hire points in a clear direction. Anthropic has been investing in long-horizon research, including work on mechanistic interpretability and scalable oversight. Adding a biologist who built a production system for protein folding suggests the company sees scientific discovery as a domain where AI can deliver measurable value, not just impressive benchmarks.
The move also reshapes the talent landscape. DeepMind has lost several key researchers in recent years. Raia Hadsell left in 2024 to join a startup. Oriol Vinyals, a lead on the AlphaStar and Gemini projects, has taken on a more advisory role. Jumper is the highest-profile departure yet. His exit raises a question: can Google DeepMind hold onto the people who built its most famous achievements, or is it becoming a training ground for Anthropic and OpenAI?
For Anthropic, the hire is a bet on depth over breadth. The company has been deliberate about growth, keeping its headcount smaller than OpenAI’s and focusing on safety research alongside capability work. Jumper fits that profile. He spent years on a single problem, not jumping between projects. He is known for insisting on rigorous evaluation and for pushing back against hype. Those instincts align with Anthropic’s public positioning.
The broader implication is for the AI industry’s relationship with science. Frontier labs are increasingly competing not just on model performance but on real-world impact. AlphaFold’s success created a template: identify a hard scientific problem, apply large-scale AI, and deliver a tool that changes how research is done. Jumper’s move to Anthropic signals that the template is portable. Other labs are now trying to replicate the formula, from Meta’s ESMFold to Baker’s work at the University of Washington. Anthropic now has the person who wrote the original playbook.
The hire is also a reminder that talent moves for reasons beyond compensation. Jumper said in his announcement that DeepMind was “a special place” and that he remains excited about its future work. He left anyway. The pull of a new mission, a different culture, or simply a new problem to solve can outweigh even the most generous retention packages. For Anthropic, that is the real prize: not just Jumper’s expertise, but the signal it sends to other researchers considering a move.
What to watch next. If Anthropic announces a biology research group in the coming months, the hire was about building a new capability. If Jumper works on alignment or safety, the hire was about applying scientific rigor to AI governance. Either way, the departure is a loss for DeepMind and a gain for Anthropic. The question for Google is whether the next generation of scientific AI breakthroughs will happen inside its labs or outside them.