On July 20, 2026, Anthropic announced a new thematic call in its AI for Science program focused on rare genetic diseases. Accepted researchers can receive up to $50,000 in Claude credits for six months. The program has two tracks: basic science partnerships and early-stage biotech work intended to accelerate clinical development. Applications close on August 2, 2026, at 11:59 p.m. Pacific Time.
Anthropic argues that rare-disease research is constrained not only by model capability but by fragmented data, small patient populations, uncertain disease mechanisms, and a long path from research findings to clinical testing. AI can help teams synthesize literature, case reports, variant data, and limited datasets, but the work still requires clinicians, patient organizations, and data scientists to validate the results. Model output is not a diagnosis or treatment recommendation.
The basic-science track includes a partnership with the Monarch Initiative. Anthropic points to Monarch’s Mondo Disease Ontology and Monarch Knowledge Graph, which reconcile disease definitions and genotype-phenotype information from multiple sources. It describes DisMech as an agent-friendly mechanistic disease classification library that lets Claude read case reports, variant databases, registry schemas, and public data, then help researchers surface mechanistic similarities for expert review.
The biotech track focuses on workflows such as organizing regulatory documentation, assessing whether a target is suitable for different therapeutic modalities, and looking for shared mechanisms across individual genetic therapies. Anthropic also names Every Cure, the Centre for Population Genomics, and the Violet Research Institute as examples of partner work involving drug repurposing, variant classification, bioinformatics pipelines, and regulatory filings. These are partner or vendor-reported use cases, not proof of clinical impact.
The structure of the program matters. It provides API credits and access to Claude Science rather than direct cash grants, while applicants remain responsible for data use, patient privacy, expert review, study design, and regulatory requirements. Anthropic also acknowledges that agents are less useful when data is sparse or poorly organized, or when the bottleneck involves insurance authorization and access to clinical infrastructure rather than information processing.
The larger signal is that AI for Science is becoming a workflow design problem. Data standards, knowledge graphs, tool permissions, traceable outputs, and expert validation need to work together. Rare diseases, with their fragmented evidence base, are a demanding test of whether agents can help researchers form better hypotheses without blurring the line between assistance and responsibility.
