Anthropic Expands AI for Science to 20 Partners

Anthropic is widening its AI for Science programme beyond individual researchers. In a post on 28 October, it named “20 scientific foundations, institutions, and research groups from around the world” it has partnered with to hand out free Claude credits for “high-impact scientific projects”. Three of them are in the UK: the Wellcome Sanger Institute, the Francis Crick Institute and Imperial College London’s Department of Life Sciences.

Two partners have since set out their plans. Imperial College London said on 29 October that its teams will use Claude on fungal disease and on a test against an unpublished discovery. Johns Hopkins Medicine said the same day that its AstroPath tumour-mapping project had been selected.

None of the three has said how many credits a partner will receive, what they are worth or how long the arrangements will last. Anthropic said “many of these collaborations are still in their early stages”.

In brief

Funders including the US National Science Foundation will pass Claude credits on to researchers.
Imperial will use Claude to design proteins against fungal pathogens and to test whether it can find an unpublished gene-transfer mechanism.
Johns Hopkins’ AstroPath team will use the credits on detailed maps of tumours and immune cells.

Who is involved in AI for Science

The post sorts the partners into three groups. Six science funders and societies will “distribute these resources to ambitious researchers”: the Alfred P. Sloan Foundation, the Simons Foundation, Schmidt Sciences, the Open Source for Science Fund run by Renaissance Philanthropy, the US National Science Foundation and CAS, a division of the American Chemical Society.

Six institutions will “use Claude credits inside their organizations”. As well as the three UK partners, they are the Space Telescope Science Institute, which runs science operations for the Hubble, James Webb and Roman telescopes; EMBL, Europe’s intergovernmental life science laboratory; and the Gladstone Institutes in San Francisco.

Six research projects get credits directly. They include FlyWire at Princeton, LEAP at Columbia, AstroPath at Johns Hopkins, gnomAD at the Broad Institute, the Tau Ceti maths library and the Centre for Population Genomics in Australia. Anthropic also pointed to earlier work with HHMI’s Janelia Research Campus on the Model Hardware Standard, “a shared specification that lets AI agents safely operate lab instruments”, which it says “is currently in research preview”.

What Imperial plans to do with Claude

Imperial says Anthropic “will provide token access to a range of research groups at Imperial” and will work “particularly closely” with its Department of Life Sciences, providing additional tools and resources.

One project targets fungal disease. A team including Professors Doryen Bubeck, Tolga Bozkurt and Nikolai Windbichler and Dr Jiorgos Kourelis has a method that Imperial says predicts “with around 90% accuracy” how a key class of fungal proteins that suppress the immune system interact with their hosts. Imperial says Claude has “demonstrated promising results in designing proteins to bind to other biological targets”, echoing Anthropic’s own protein-binder work on scorpion toxins. The team plans to design and test proteins against fungal pathogens. The long-term aims are crops with built-in disease resistance and new treatments for fungal infections in people.

A second test is stricter. Dr Tiago da Costa’s group recently found a gene-transfer mechanism in its lab that has not been published. It will check whether Claude can identify it independently, using only evidence that was public before the discovery. “The real promise of AI is not simply to make science faster, but to help us think differently and uncover biology that would otherwise remain hidden,” he said.

Imperial’s Provost, Professor Peter Haynes, said the university wants to work “with the companies building these tools, not simply using them at a distance.”

AstroPath at Johns Hopkins

AstroPath was built by astrophysicist Alexander Szalay and pathologist Janis Taube. It bases its data infrastructure on the one built for the Sloan Digital Sky Survey and maps where cancer cells, immune cells and other cells sit within tumours. The Mark Foundation for Cancer Research has invested “more than $10 million” in the work.

Johns Hopkins says outlining tumour cells by hand, once painstaking, “can now be completed in minutes using AI-assisted tools”. Taube stressed the limits. “These are supervised tools,” she said. “We still oversee the process, but AI allows us to generate datasets much faster and with far more detail than we could by hand.” Szalay and Taube are also exploring a possible company for AI-powered tumour analysis and biomarker tests.

How AI for Science changes who decides

The programme started in May 2025 with credits for individual researchers, and Anthropic says it has since supported “hundreds of projects”. In August it added a team plan for scientists, and in October it committed $150 million over three years to the US Genesis Mission.

Working through funders and institutes moves the choice of projects away from Anthropic. The NSF or the Simons Foundation can put credits into their own grant rounds, and Imperial can hand them to its own groups. That is a cheaper way to reach thousands of researchers. It also ties Claude into the way science is funded, which matters to Anthropic as it competes for researchers, as its published claims about Claude’s research abilities and drug-development work show.

The partner statements make the offer easier to judge. Imperial’s da Costa test is a useful design: if Claude finds a mechanism the lab already knows but has not published, that is evidence of reasoning rather than recall. If it does not, the result is just as informative. Hopkins, by contrast, is describing faster analysis of data it already holds, under human supervision.

It is a different model from Meta’s grants to five US universities, which came with cash, and from the European Commission’s AI in Science Enablers initiative. Anthropic’s offer is computing time on its own models, with no published value, so it cannot yet be compared with a typical grant. For now these are partnerships and plans, not discoveries. More on AI in research is in our Tech for Good hub.