Cui On October 28, Anthropic expanded its AI for Science program beyond individual researchers. Twenty foundations, institutions and research projects now receive free Claude credits to distribute or use internally.
Why It Matters
The program started in May 2025 by giving credits to individual scientists. Hundreds of projects later — across biology, medicine, physics, chemistry, climate and mathematics — Anthropic is moving up a level: funders and institutions multiply reach far more than one-by-one grants.
Three Tiers of Partners
- Funders and societies distribute credits to researchers: Sloan Foundation, Simons Foundation, Schmidt Sciences, Open Source for Science Fund (Renaissance Philanthropy), the US National Science Foundation, and CAS (American Chemical Society).
- Institutions use credits in-house: Space Telescope Science Institute (Hubble, Webb, Roman), Wellcome Sanger Institute, Francis Crick Institute, Imperial College London Life Sciences, EMBL (including EMBL-EBI) and Gladstone Institutes.
- Research projects get credits directly, such as FlyWire (Princeton, the complete adult fruit-fly brain wiring map), LEAP (Columbia, ML plus physics for climate models) and AstroPath (Johns Hopkins, cancer research).
The Engineering Angle
Two earlier collaborations show where this is heading:
- Model Hardware Standard (MHS) with HHMI Janelia: a shared specification that lets AI agents safely operate lab instruments. Currently in research preview.
- Allen Institute: teams of agents analyzing large open brain, cell and immune datasets, turning months of manual analysis into hours.
MHS is the interesting one. Agents controlling physical instruments need a safety contract, much like MCP gave agents a tool contract. Expect standards like this to matter more than the credits.
Public Goods
Anthropic expects many projects to yield open datasets, open-source software, open standards and shared libraries — benefits that outlast the credits.
Takeaway
This is less a product launch than a distribution strategy: embed agents in the institutions that own the data and instruments. The open question is how outputs are validated at scale. Credits buy compute; trust in agent-generated findings still has to be earned.
Source: Anthropic
Click to load Disqus comments