There’s a real, urgent need for help in rare disease research. Anthropic launched a call within its AI for Science program to fund projects that use its API and Claude in two tracks: basic research and early-stage biotech. Why should you care if you work with clinical data, a patient NGO, or a biotech startup? Keep reading.
What the program offers
Anthropic will award up to USD 50,000 in Claude credits per project, distributed over six months. The goal is to build a community of teams that share findings, data and methods to speed discoveries in rare diseases.
There are two main tracks:
- Track 1: collaborative basic research between clinicians, patient organizations and data scientists.
- Track 2: biotechs and teams aiming to accelerate clinical and regulatory development of therapies for rare diseases.
The deadline to apply is August 2, 2026 at 11:59 PM PST. Awardees will be able to use Claude Opus and other approved biology models; some projects that trigger detectable risks by bio-classifiers may request exemptions.
Why it matters for rare diseases
Did you know rare diseases collectively affect hundreds of millions of people? Anthropic cites about 400 million people and more than 7,000 diseases, while other sources push that number toward 10,000. The problem isn’t just low incidence: data is fragmented, patient registries are small, and the literature is scattered.
The idea is simple: AI can organize, synthesize and spot patterns that would take human teams far longer to see. For example, a system like Claude can read case reports, variant databases and registry schemas to suggest mechanistic similarities between diseases that no one had connected before.
Track 1: basic research and shared resources
This path aims to boost collaborations with initiatives like the Monarch Initiative. Monarch already provides tools such as the Mondo ontology and a knowledge graph that integrate genotype-phenotype data across species.
A key contribution is DisMech, an agent-oriented library that helps mechanistically classify diseases so models like Claude can find links and propose hypotheses experts can test. Results from this track will be published publicly on Monarchinitiative.org and there will be community efforts like hackathons.
Examples of valid projects in this track:
- Propose and prioritize mechanistic links between rare diseases that share genes or pathways.
- Curate and summarize patient-organization data to improve natural history studies.
- Build evaluations that measure how well models solve specific rare-disease tasks and where they fail.
Track 2: biotech and clinical acceleration
This track is aimed at teams that want to compress stages of therapy development. In practice, Anthropic suggests AI can help to:
- Draft and review regulatory documents, such as IND sections and CMC modules, to cut weeks or months of manual work.
- Select therapeutic strategies by analyzing whether a target is addressable by different modalities: small molecules, antibodies or gene therapies.
- Identify shared mechanisms that enable basket trials instead of separate INDs per patient.
Concrete examples Anthropic cites:
- Rationalize starting doses from sparse data using PK/PD modeling synthesis and precedents.
- Mine natural histories and case reports to find biomarkers and practical endpoints for N-of-1 studies.
- Automate regulatory drafts to compress dossier preparation to days of expert review.
Groups are already using Claude in this space: Every Cure looks for drug repositioning opportunities; the Centre for Population Genomics is building systems that draft variant classifications for review; and the Violet Research Institute uses Claude in bioinformatics flows and regulatory documentation.
Limitations and responsibilities
The call is useful, but it’s not a magic wand. Anthropic acknowledges that:
- AI depends on the quality and quantity of data: if data are scarce or poorly organized, results will be limited.
- There are areas AI can’t solve alone, like access barriers to diagnosis, insurance approvals, or constraints in manufacturing and safety.
- Human scrutiny and experimental validation are necessary; model outputs must be reviewed by experts.
That’s why the program encourages collaboration between clinicians, patients and data experts, and asks for transparency about model failures and limitations.
How to apply
If you have a project that fits either track, prepare a clear proposal: objective, available data, team, and how you’d use Claude credits. Apply before August 2, 2026 at 11:59 PM PST. Projects that interact with bio-classifiers can request exceptions case-by-case.
If you work with patient organizations or in a small startup, this call could give you the technical and community boost needed to test ideas that otherwise wouldn’t get support.
We’re at a point where AI can help connect pieces that have been scattered for decades. It doesn’t replace the scientific method, but it can accelerate hypothesis generation, evidence compilation and drafting critical documents. Want to explore how your project should fit into this ecosystem? This call is a door to try.