Anthropic Starts Its Own Drug R&D — Targeting the Diseases Big Pharma Ignores

Anthropic is no longer content selling AI tools to drug companies. At a briefing this month, the company announced it would start discovering drugs itself — specifically for neglected diseases that big pharmaceutical companies have written off as commercially uninteresting.

The new initiative runs alongside Claude Science, a research workbench Anthropic launched this month for scientists. Claude Science pulls scattered research tools and datasets into a single environment — data analysis, chart generation, research assistance all live in one place. The company showed off one case study during the event: a researcher at UCSF used Claude Science to spot a viral contamination in an experiment — in minutes. The same problem had gone undetected by the research team for a year.

Anthropic claims the system can analyze 100 rare genetic diseases in under an hour and screen out 32 directions worth pursuing computationally. That speed is the selling point. But the company has not disclosed which specific diseases it will target first, nor has it clarified whether it will take a promising candidate through clinical trials itself or partner with established players for the long haul of animal tests, human trials, and eventual manufacturing.

This puts Anthropic in a rare position among major AI labs. OpenAI, Google, Amazon, and Google DeepMind’s Isomorphic Labs all offer AI platforms for life sciences. But very few have said they will touch the wet-lab side of drug development themselves.

Novartis CEO Vas Narasimhan, speaking at the same event, put the challenge in perspective. A new drug takes about 12 years from discovery to approval, he said. Information processing and operations account for about 40% of that timeline — and he believes AI could compress those stages significantly, potentially cutting the full cycle to seven or eight years. But the biological validation — animal studies, human clinical trials — cannot be sped up the same way. He also predicted AI could lift drug success rates from roughly 8% to about 16%, a meaningful improvement that still leaves most candidates failing.

Academics were more cautious. Namshik Han, a Cambridge professor and co-founder of AI biotech firm CardiaTec, noted AI can participate in nearly every stage of drug development — finding compounds, optimizing molecules, analyzing data, supporting clinical work. But clinical validation remains the bottleneck. Matthew Todd, a drug discovery professor at University College London, pointed out that no drug designed entirely by AI has completed all clinical trials and received FDA approval. AI accelerates candidate discovery and research efficiency, he said, but it cannot replace the pipeline itself. Frank von Delft, an Oxford structural biology professor, added that AI has not made experimental work optional. Candidates still need to pass toxicity, safety, and efficacy hurdles — all of which demand capital, specialized personnel, and often years of work.

Anthropic has been staffing up for this for more than a year. The company has hired across biology and life sciences roles, built a wet lab, and recruited from large pharma companies and research institutions. The infrastructure is coming together.

The broader trend is real. AstraZeneca, Novo Nordisk, and GSK have all integrated AI into parts of their pipeline — drug screening, molecular design, and clinical data analysis among them. The question is not whether AI belongs in pharma — it clearly does — but whether an AI company can compete with decades of pharmaceutical expertise when the experiments start running. The answer to that question is years away.