On July 29, 2026, the Financial Times reported that Google DeepMind had dissolved AlphaFold's dedicated team and moved most of the researchers behind the original paper to different roles over the past year. DeepMind reportedly confirmed the personnel moves. According to the newspaper, the research organization that led to the 2024 Nobel Prize in Chemistry has effectively disappeared, with core researchers including John Jumper heading to Anthropic.
However, this does not mean the AlphaFold technology or service itself is being discontinued. The official page still lists a database containing over 200 million structure predictions, the AlphaFold Server, and AlphaFold 3 for academic use. What has shifted is the organizational unit behind building scientific AI. DeepMind is moving away from a model of dedicated teams tackling single hard problems, toward one centered on Gemini as a command hub coordinating multiple agents and specialized models.
What the AlphaFold "Dissolution" Report Says
According to the FT, nearly a quarter of the authors credited on the original AlphaFold paper who were full-time at Google DeepMind at the time of writing have already left the company. Most researchers who remain internally have moved into Gemini-related work. Others now handle individual projects ranging from life sciences to fusion energy. Some reportedly moved to Isomorphic Labs, another Alphabet subsidiary. Since DeepMind has not disclosed the number of people affected or its current organizational chart, the denominator behind "nearly a quarter" and the destinations of all those involved cannot be verified.
The most symbolic figure in this outflow is John Jumper, who led AlphaFold. On June 19, Jumper announced on his own X account that he was leaving DeepMind after about nine years, and would join Anthropic after taking a break. The 2021 Nature paper states that Jumper and Demis Hassabis led the research. In 2024, the two shared a quarter each of the Nobel Prize in Chemistry for their contributions to protein structure prediction.
On June 24, Bloomberg reported that Jonas Adler and Alexander Pritzel, also authors on the same Nature paper, were expected to join Anthropic as well. The FT further reported that Jumper and Adler had moved earlier this year to an internal team strengthening AI coding. The reasons behind the three departures and their roles at Anthropic have not been disclosed. From this alone, one cannot draw conclusions about compensation, research autonomy, or internal culture as causes.
From a Small Team in 2016 to Gemini as Command Hub
AlphaFold emerged from DeepMind's research approach of concentrating talent on a single, measurable hard problem. According to the official timeline, a small team was set up in March 2016 to tackle protein structure prediction. After taking first place at CASP13 in 2018, the team expanded, and by CASP14 in 2020 it had far outpaced the runner-up. The 2021 Nature paper reported a median backbone accuracy of 0.96 Å, compared to 2.8 Å for the next-best entry.
The conditions for success were clear. With CASP providing an external evaluation using unknown protein structures, the dedicated team could keep improving a single metric. Speaking to the FT, Pushmeet Kohli, who leads AI for Science, explained that over the past nine years the focus had been on setting concrete goals for individual "grand challenges," but that the strategy has since evolved.
The new design took concrete shape with Co-Scientist, which began experimental availability in May 2026. Built on Gemini, multiple agents iterate through generating hypotheses, critiquing them, ranking them, and refining them. Beyond literature search, they reference databases such as ChEMBL and UniProt, and in selected research collaborations, they also call on specialized models like AlphaFold as tools.
While AlphaFold was a model built to solve a single hard problem, Co-Scientist aims to be a coordination layer spanning the entire research pipeline. Rather than discarding specialized models, it reuses them from a common agent foundation. This shift is consistent with the FT's report that researchers have been redeployed from Gemini-related work to fields ranging from life sciences to fusion energy.
Anthropic Is Also Moving to Capture the Scientific Research Workflow
Claude Science, which Anthropic released on June 30, restructures scientists' working environment from a direction similar to DeepMind's. A general-purpose coordinating agent draws on over 60 skills and connectors covering genomics and single-cell analysis, as well as proteomics, structural biology, and cheminformatics. It also calls on specialized agents built by users, while a separate review agent checks citations and computations.
Both companies' product visions have moved away from researchers feeding queries into individual models one at a time. DeepMind has turned AlphaFold into a tool orchestrated around Gemini and Co-Scientist, while Anthropic bundles scientific databases, computational environments, and specialized agents around Claude. The competition has expanded from the accuracy of protein structure prediction to the entire research pipeline—from forming hypotheses, to running computations, to producing verifiable outputs.
The hiring of Jumper and others coincides with the timing of Anthropic's beta release of Claude Science. Even so, it has not been confirmed that the three will work on Claude Science or biology models specifically. What is confirmed is only that researchers who worked on both AlphaFold and Gemini are moving to a competitor that has launched a general-purpose working foundation for science.
AlphaFold's Legacy Remains Across Three Operational Layers
Even as the dedicated team has been dissolved, as the FT reported, the assets AlphaFold produced have not disappeared. They are now divided across three layers with different roles.
| Operational Layer | Current Assets | Role |
|---|---|---|
| Public research infrastructure | AlphaFold Database, Server, and AlphaFold 3 for academic use | Broadly providing structure prediction to scientists |
| DeepMind's common foundation | Gemini, Co-Scientist, AlphaFold integration | Coordinating from hypothesis generation to use of specialized models |
| Alphabet's drug discovery business | Isomorphic Labs' IsoDDE | Connecting structure prediction to drug design and candidate development |
The scale of the public research infrastructure is already substantial. According to DeepMind, by November 2025, AlphaFold had been used by over 3 million researchers in more than 190 countries, with over 30% of related research addressing disease understanding. Separate from whether the team continues to exist, whether the database and server are kept updated and accessible to researchers will determine its public value.
On the commercial side, Isomorphic Labs announced in February 2026 that its drug discovery engine, IsoDDE, had more than doubled AlphaFold 3's accuracy on a generalization benchmark for difficult protein-ligand structure predictions. This is a limited comparison based on the company's own technical report. In May, the company announced a $2.1 billion Series B funding round, stating it would use the funds to advance drug candidates into clinical stages.
The dissolution of AlphaFold's dedicated team, as reported by the FT, is not evidence that DeepMind has abandoned science. But having given up an organization that concentrated on a single hard problem since 2016, it remains to be seen through actual results whether a general-purpose foundation can produce externally verifiable achievements on par with AlphaFold. Whether Co-Scientist yields discoveries that withstand external verification, whether AlphaFold's public assets continue to be updated, and what the three researchers moving to Anthropic ultimately build—the success or failure of this organizational change can only be judged as these three points become clear.