John Jumper, the Nobel Prize-winning scientist who led the creation of AlphaFold and cracked one of biology’s oldest problems, is leaving Google DeepMind for Anthropic, according to people familiar with the matter. The move, expected to be announced this week, hands the AI safety company one of the most decorated researchers in the field, and deepens the questions about talent retention inside Google’s London-based research unit.
Jumper shared the 2024 Nobel Prize in Chemistry with DeepMind co-founder Demis Hassabis and University of Washington researcher David Baker for work on protein structure prediction. His AlphaFold system, released in its second form in 2020, solved a problem that had frustrated biologists for half a century: predicting the three-dimensional shape of a protein from its amino acid sequence. The tool’s database now holds predicted structures for more than 200 million proteins, and researchers in nearly every country have used it, from drug discovery labs to agricultural science departments.
His departure follows that of Noam Shazeer, the co-lead of Google’s Gemini model program, who left the company earlier this year. Shazeer, a co-author of the Transformer paper that underpins modern AI, had returned to Google in 2024 after founding Character.AI and had been positioned as a central figure in the Gemini push. Two exits of that stature inside two years have rattled researchers inside DeepMind, according to people familiar with the group’s internal discussions, and have prompted Google executives to offer expanded retention packages to senior scientists.
Anthropic has made no secret of its recruiting strategy. The company, founded by former OpenAI researchers, has framed itself as the laboratory most committed to understanding and controlling the models it builds, and it has spent heavily on interpretability research, the effort to see inside neural networks. That positioning, combined with access to large computing budgets and a model roadmap that has kept pace with the frontier, has made it a magnet for researchers who want both scale and a distinctive scientific agenda, several AI academics said.
For Jumper, the pull appears to be scientific as much as financial. Colleagues described him as increasingly interested in the biological applications of frontier models — using large language models to reason about molecular systems, design proteins and accelerate drug discovery — and Anthropic has been building a life-sciences effort around exactly those questions. A person close to the company said Jumper will join its science division with a mandate to apply frontier models to biological problems, rather than to safety research narrowly defined.
The loss for Google is symbolic as well as practical. AlphaFold was the company’s most celebrated scientific achievement of the past decade, a demonstration that its AI could do more than predict the next word. Watching its lead author walk out the door hands rivals a talking point about Google’s ability to keep its best people, and hands Anthropic a figure whose name carries weight in academic circles where AI companies increasingly recruit.
The broader pattern is bigger than any single hire. Frontier AI research has become a market where a handful of laboratories — Google DeepMind, OpenAI, Anthropic, Meta and a small set of startups — compete for a talent pool measured in the hundreds of world-class researchers. Compensation packages run to eight figures, compute allocations are negotiated like contracts, and the freedom to choose research problems has become a recruiting weapon. Anthropic’s pitch, that safety work will not be sacrificed for shipping deadlines, has resonated with a specific slice of that pool, analysts who track AI hiring said.
Google is not standing still. The company has restructured DeepMind’s reporting lines, expanded research budgets and promoted younger scientists into leadership roles, according to people familiar with the plans. But the departures have already had an effect: Gemini’s leadership changed hands mid-project, and some external observers now describe Google’s frontier model program as having lost two of its most visible technical advocates.
Anthropic, for its part, is assembling a roster that now includes a Nobel laureate, and the symbolism is unlikely to be lost on investors or on the regulators watching the industry. The company has said repeatedly that its advantage lies in doing AI differently; hiring the scientist who did biology differently is the strongest evidence yet that the pitch is landing.
The scale of what AlphaFold unlocked is hard to overstate. Before the system, a researcher hoping to determine a protein’s structure faced months of laboratory work, and many proteins resisted every technique available. AlphaFold delivered predictions in minutes, and the freely available database that followed became a standard tool in thousands of laboratories, from teams hunting new antibiotics to groups studying how plant proteins respond to drought. When the Nobel committee announced the prize in 2024, it described the work as solving a challenge that had stood for half a century.
Anthropic’s scientific ambitions have grown alongside its commercial ones. The company has said it is building toward systems that can assist research well beyond software, and the hiring of a structural biologist signals that those bets are becoming concrete. People familiar with the company’s plans said Jumper’s group will be free to set its own agenda, a freedom that DeepMind’s size and its obligations to Google’s product teams have made harder to grant.
Neither Google DeepMind nor Anthropic commented on the move, which has not yet been publicly announced. People familiar with the matter said Jumper will begin at Anthropic in the coming months. His first project, they said, is likely to be one that combines the two fields he has spent his career bridging: machine learning and molecular science.
