Anthropic moves from Claude Science tooling toward its own drug candidates

Anthropic is pushing Claude Science beyond the role of a research assistant. According to The Verge’s July 3, 2026 report, the company said at its AI for Science briefing that it wants to develop drugs of its own, with an initial focus on neglected diseases. That is the more consequential part of the story: a frontier AI lab is testing whether it can move from selling tools to scientists into generating therapeutic candidates itself.

The product backdrop is Claude Science, Anthropic’s new workbench for scientific research. The official launch framed the system as a way to pull literature review, datasets, code execution, figure generation, compute access, and validation into one environment. That already matters for labs juggling PubMed, notebooks, domain databases, and HPC workflows. The follow-up signal is sharper: Anthropic appears to see the workbench not only as software for external biopharma customers, but also as infrastructure for its own discovery programs.

The Verge reported that Eric Kauderer-Abrams, Anthropic’s head of life sciences, said the company would focus on treatments for neglected diseases. The company has not said which diseases it will target first, what it would do with a promising candidate, or whether it would partner for animal studies, clinical trials, manufacturing, or commercialization. Those missing details are not footnotes. Drug discovery only starts with a candidate. It still has to survive biology, toxicity, formulation, dosing, manufacturing, regulation, and human trials.

That is why the expert caveats in the report matter. Namshik Han of the University of Cambridge described “AI drug discovery” as a broad term that can cover almost every stage of the process, from molecule search to clinical and manufacturing support. Oxford’s Frank von Delft warned that AI models have not made experiments unnecessary. A molecule that looks plausible in software still has to prove efficacy, safety, stability, and deliverability in the physical world.

Still, Anthropic’s move changes the industry map. OpenAI, Google, Amazon, and specialist AI biotechs all want science and pharma workloads. But a frontier model lab that develops its own drug candidates becomes something different from a vendor. It could sell productivity software to pharma customers while also competing, at least in selected disease areas, for the value created by discovered assets.

The next evidence to watch is concrete rather than rhetorical: target disease areas, wet-lab partners, candidate ownership, preclinical milestones, and whether Claude Science’s audit trail can connect model-generated hypotheses to reproducible experimental results. AI can speed parts of the search, but the market will judge Anthropic’s drug push by molecules that reach real validation, not by workflow demos.