Anthropic said it will move past selling AI tools to pharmaceutical companies and start developing drugs itself, with an early focus on neglected diseases. The company, known for the Claude model, announced the plan alongside Claude Science, a new AI workbench built for scientific research.
Claude Science was introduced before the drug push
According to The Verge, Anthropic first used its “The Briefing: AI for Science” event to present Claude Science. The product is designed to bring research tools and large datasets into one environment and generate complex visualizations automatically, with the goal of speeding up scientific discovery and medical intervention. The bigger surprise came later in the event, when the company outlined its ambitions in drug development.
Eric Kauderer-Abrams, who leads Anthropic’s life sciences group, said the company does not want to stop at providing AI systems to drugmakers. It wants to build medicines on its own. He said the initial research effort will target diseases that draw limited commercial interest and often receive less attention from traditional pharmaceutical firms. That shift could put Anthropic in an unusual position, selling Claude to major drug companies while also competing with them in selected disease areas.
The company is building beyond software
The report said Anthropic has been hiring biologists and has already set up dedicated wet labs. That suggests the company is trying to connect model-driven molecule discovery with physical validation in the lab. It is a rare posture among frontier AI developers. OpenAI, Google, and Amazon have all launched life sciences tools, but few have publicly said they plan to develop physical drugs themselves.
Researchers say the path to market remains long
Academic experts interviewed in the report gave a much colder assessment of how fast AI can change drug development. Frank von Delft, a professor at Oxford, said AI is still far from replacing real-world experiments. Even if a model predicts a promising new molecule, that candidate still has to go through tests for toxicity, stability, and efficacy in living systems. Preclinical and clinical work remain expensive, and failure rates are high.
Matthew Todd, a professor of drug discovery at University College London, added that AI has real potential in testing ideas for new medicines more quickly, but chemistry inside biological systems is extremely complex. Large gaps in medical knowledge remain, and algorithms cannot remove those gaps on their own.
At present, the report noted, no drug designed entirely by AI has completed full clinical trials and won FDA approval. It also said the path from discovering a new target to reaching the market still takes at least 10 years. Anthropic has not disclosed the disease target for its first candidate drug, and it has not said whether later-stage clinical testing and manufacturing will be handled internally or in partnership with established pharmaceutical companies.