DeepMind Disbands AlphaFold Team as Jumper Departs for Anthropic

The team behind AlphaFold, the system that won Demis Hassabis and John Jumper the 2024 Nobel Prize in Chemistry, has been dissolved, Engadget reported, citing Financial Times reporting. Most of the original authors were quietly reassigned over the past year to Gemini-related projects, Isomorphic Labs, or research areas like enzyme design, nuclear fusion, and genomics. Almost a quarter of the researchers have left Google DeepMind entirely.

The most visible departure is Jumper himself, a VP and engineering fellow, who left in June to join Anthropic. Colleagues Jonas Adler and Alexander Pritzel are reportedly following him. Those are named authors on the AlphaFold papers walking into a direct competitor, not a rounding-error headcount move.

DeepMind is framing this as a pivot rather than a retreat. Pushmeet Kohli, VP of Research and founder of the AI for Science team, said 'the strategy has evolved,' and the company no longer wants to organize scientists around a single problem like protein folding. It wants Gemini-powered systems that can assist researchers across disciplines and eventually automate parts of the scientific process. An official statement said DeepMind remains 'incredibly proud' of AlphaFold's impact and that its researchers 'continue today to drive scientific and technological advances across Google, Google DeepMind, and Isomorphic Labs.'

The honest caveat is that the reporting doesn't tell you who maintains the AlphaFold Protein Structure Database, which houses more than 200 million predicted structures used by working biologists, or whether future model updates are on any roadmap. It also doesn't detail the terms of the Anthropic hires. Take the 'team disbanded' framing as reported, not as a full obituary. Many of these people are still inside Google, just pointed at different targets.

The forward-looking read is that Anthropic has just acquired a serious slug of biology-adjacent talent at the exact moment frontier labs are racing to prove their models can do real science. Google's bet is that a general-purpose Gemini agent will eventually beat a bespoke specialist team on the next protein-folding-scale problem. That is a much bigger claim than 'we're refocusing resources,' and it will take more than a press line to settle.