Anthropic adds 20 partners to AI for Science, but the post names no credit amounts or results

Anthropic's AI for Science program added 20 partners on October 28, 2026, including the US National Science Foundation, the Simons and Sloan foundations, Schmidt Sciences, the Wellcome Sanger Institute and the Francis Crick Institute. The announcement is mostly a list of names and a promise of free Claude credits, but one line in it points at something larger: a standard for letting AI agents run physical lab equipment.

Who joined

Anthropic launched AI for Science in May 2025. It gives researchers on high-impact projects free Claude credits, and Anthropic says it has supported hundreds of projects across several fields. The October 28 post adds 20 partners in three groups, as the company lists them.

Funders and societies (six): the Alfred P. Sloan Foundation, the Simons Foundation, Schmidt Sciences, the Open Source for Science Fund at Renaissance Philanthropy, the US National Science Foundation and CAS, a division of the American Chemical Society.

Institutions (six): the Space Telescope Science Institute, the Wellcome Sanger Institute, the Francis Crick Institute, Imperial College London's Department of Life Sciences, EMBL and the Gladstone Institutes.

Research projects (six): FlyWire at Princeton, LEAP at Columbia, AstroPath at Johns Hopkins, gnomAD at the Broad Institute, Tau Ceti and the Centre for Population Genomics. Earlier partners named in the post include HHMI's Janelia Research Campus and the Allen Institute.

The post opens with its own framing: "AI for Science began by supporting individual scientists." The expansion shifts the model from handing credits to individuals toward partnering with the bodies that fund and run research.

What the post does not say

The page does not say how many credits each partner receives, what they cost Anthropic, or what outputs partners must produce. It gives no results from the new partners, since they have only just joined. The one number about impact, "hundreds of projects", is Anthropic's own and carries no list of outcomes.

That makes the announcement a distribution and relationship story, not a capability one. Putting Claude into the NSF, two large philanthropies and several of Europe's biggest genomics and imaging institutes is a strong position for a lab that wants to be the default research assistant. It is also a customer-acquisition channel, and the credits are free for now.

Some context on scale helps read the names. The post describes the NSF as a roughly $9 billion federal agency that supplies about a quarter of federal basic research support at US colleges and universities, and says gnomAD holds DNA data from more than 800,000 people. Those are descriptions of the partners, not of what Claude will do for them.

The part that is not a press release: Model Hardware Standard

Buried in the announcement is the Model Hardware Standard, or MHS, developed with HHMI Janelia Research Campus. The October 28 post calls it "a shared specification that lets AI agents safely operate lab instruments" and says "MHS is currently in research preview."

It is not new on October 28. Coverage of the August 27, 2026 opening of the research preview describes a standard driver paired with a reference file that states what a machine can measure and change and which safety limits an agent must obey. The spec is model-agnostic and reachable through MCP, a command line or code and APIs, so it is not tied to Claude. Reported vendor support includes AWS through Strands Robots, with Automata, Doosan, MBF Bioscience, QIAGEN, Tecan and Universal Robots also listed, while Danaher is described only as exploring a tie-in. Anthropic plans to open-source MHS with safety evaluations later and is running a waitlist.

This matters because it moves AI from analyzing data to acting on the world. A model that can drive a microscope or a liquid handler can run experiments, and an experiment that goes wrong can ruin samples or damage equipment. The safety design, limits in a machine-readable file that the agent must respect, is sensible, but the public detail is thin. The spec's safety evaluations have not been published.

How this connects to Anthropic's other science claims

Anthropic has been making stronger science claims elsewhere. We looked at its gene editing discovery claim and what it means for lab oversight, and Anthropic has described preclinical results from six drug development partners. The AI for Science expansion is the supply side of those claims: credits and access that bring more scientists, and more data, into the Claude ecosystem.

The race angle is simple. If the labs believe AI will speed up scientific discovery, the winners will be those embedded in how research is funded and run. A grant-making body that standardizes on one lab's tools shapes which models get used for years.

Our take

Treat this as a credible expansion with unquantified substance. The partner list is real and checkable on the post. The impact claims are not yet evidence. We would watch three things: whether Anthropic publishes the credit volumes and outcomes, whether MHS is open-sourced with the promised safety evaluations, and whether the funders disclose any usage terms, such as data-sharing or independence safeguards, when public money is involved.

For researchers, the practical step is to apply if your project fits, read the terms on data and publication, and keep your own evaluation of any result a model produces. A free credit is not a validated finding, and a lab announcing who it works with is not the same as a lab showing what the work produced. The first results from these 20 partners, when they arrive, will be the real test of the program.