Anthropic has opened applications for a rare genetic disease research programme that will give accepted researchers and early-stage biotechnology companies up to US$50,000 in Claude credits over six months. The call is the first topic-focused round within Anthropic’s broader AI for Science programme and is accepting applications until 2 August 2026 at 11:59 pm Pacific time.
The programme has two tracks. One supports basic-science work that brings together clinical researchers, patient organisations and data scientists. The other focuses on early-stage biotechnology teams seeking to accelerate parts of clinical development, including evidence review, therapeutic strategy selection and regulatory documentation.
A focused extension of AI for Science
Anthropic says its existing AI for Science initiative has supported work ranging from drug repurposing to quantum simulation. The company found that grantees working on related questions benefited from sharing methods and has therefore moved towards thematic calls that can create a more connected research community.
Rare diseases present an unusual information problem. Individual conditions affect small populations and are often studied separately, while relevant evidence is distributed across case reports, registries, variant databases and incompatible classification systems. Anthropic argues that language models can help researchers synthesise literature, extract information from limited datasets and identify mechanisms that may be shared across conditions.
Basic science track
The first track is being developed with the Monarch Initiative, an international consortium that builds standards and knowledge resources for rare-disease diagnosis and mechanism discovery. Its resources include the Mondo Disease Ontology, which reconciles definitions from multiple classification systems, and the Monarch Knowledge Graph, which connects genotype and phenotype information across species.
Anthropic also points to DisMech, an agent-oriented mechanistic disease classification library. Proposed projects could use Claude with these resources to rank links between diseases that share genes or pathways, curate patient-organisation data for natural-history studies, or build evaluations that measure where models succeed and fail on rare-disease tasks. Outputs from the basic-science track are expected to be made public through the Monarch Initiative.
Biotechnology and clinical development track
The second track is aimed at biotechnologists and early-stage companies. Anthropic suggests Claude could assist with tasks such as comparing therapeutic modalities, analysing whether a target is druggable, mining natural-history evidence for useful biomarkers and drafting or cross-checking regulatory material for expert review.
Accepted teams can use credits with Claude Opus and other generally available models approved for biology. Anthropic says projects that encounter its biological-use classifiers may be considered for exemptions, indicating that access will still operate within a controlled review process rather than as unrestricted model use.
The company cites existing grantees as examples. Every Cure is using Claude to search for drug-repurposing opportunities, while Australia’s Centre for Population Genomics is developing a system that drafts genetic-variant classifications for expert assessment. Anthropic also describes work at the Violet Research Institute involving bioinformatics, experimental analysis and regulatory filings for ultra-rare diseases.
Important limits remain
Anthropic acknowledges that AI cannot compensate for missing, poor-quality or poorly organised data. It also notes that language models cannot resolve access barriers such as insurance authorisation, diagnostic infrastructure or manufacturing and safety-testing constraints. Those qualifications are important because a faster document workflow is not the same as a faster or safer therapy.
For research organisations, the practical value of the call is access to substantial API credits, Claude Science and a peer community focused on related problems. Applicants will still need appropriate domain expertise, governance and human review, particularly where outputs could affect diagnosis, trial design or regulatory decisions.
Application considerations
Applicants should define a question that can be evaluated within the six-month credit period and explain how clinicians, patients, biologists or regulatory specialists will review the model’s contribution. A project that merely produces more text is less useful than one with a measurable outcome, such as better retrieval of supporting evidence, faster preparation of a reviewed dossier or a benchmark that clearly records failure cases.
Data governance will also shape what is feasible. Rare-disease datasets may contain identifiable health information and can remain sensitive even after obvious fields are removed because the affected population is small. Teams should establish lawful access, retention limits, audit trails and a clear separation between research assistance and clinical decision-making before sending material to any hosted model.
The grant covers Claude usage rather than the full cost of a research programme. Staff time, data preparation, validation, compute outside Anthropic’s service and any laboratory or clinical work remain the responsibility of participating organisations. The programme is therefore best understood as targeted technical support, not as a conventional unrestricted research grant.
Anthropic says successful projects may use generally available models approved for biology, but model availability can change during a six-month programme. Applicants should document the model and settings used for every result and plan for reproducibility if a newer version appears. Public outputs from the basic-science track should distinguish generated hypotheses from evidence that has been independently validated.