Anthropic tests Claude across drug development, from labs to trials

Where Claude enters the process

Drug development involves more than designing a molecule. Teams must choose a target and treatment approach, produce a molecule with the desired properties, prepare it for reliable manufacturing, assemble regulatory documents, and plan and run clinical trials.

Anthropic says bottlenecks appear throughout that process, so it is studying Claude on real programs across different medical fields and stages. The company says the early projects have already helped identify improvements to its models. It hopes to work with more biotech companies and embed Claude more deeply in their workflows.

The first set of examples includes work on food allergy, cancer, rare childhood diseases, and clinical-trial planning:

Generate Biomedicines used Claude Code to propose shorter genetic instructions for CAR T-cell therapy. One version was about 70% shorter than the standard. In laboratory tests, immune cells produced about 10 times more receptors, killed tumor cells more effectively, and remained active longer.

Divergence Labs asked Claude to turn its drug-design goals into a development plan, then generate about 35,000 antibody variants in roughly a month. Early tests found three designs that bound to IgE and 20 that helped detach it from immune cells. Three of those 20 appeared to outperform a standard drug at detaching IgE.

BigHat Biosciences and an Anthropic researcher used Claude to improve two antibody-fragment candidates. Over about 10 weeks, the lab made and tested about 1,010 variants. Compared with the original molecules, the best candidates bound to their target more than 10 times as tightly. One remained dissolved at a temperature 18 °C higher; another clumped more than 100 times more slowly.

Enveda connected manufacturing records to Claude to flag possible problems. In 1,061 pages of documents, Claude found all 11 issues identified by human reviewers and seven they had missed. Human review time fell by 89%, from more than 25 hours to under three.

Bristol Myers Squibb is using Claude to assess early-stage trial plans and estimate their chances of success. The partners plan to expand the work in 2027. BMS says one prompt produced a substantive initial analysis and assessed development options in a task that otherwise would have taken several days.

The work behind the numbers

Some of the examples are about research; others are about the labor and infrastructure around it. Enveda’s document review and The Column Group’s new company, Pleiades, address the operational burden of running drug programs. Pleiades is building tools and workflows for Claude, starting with a skill to assess whether particular drug targets are worth pursuing. It has been tested on 45 target-disease pairs.

Anthropic has also acquired Mithril Medicine, a one-person company developing treatments for two rare forms of muscular dystrophy. Founder Cody Trumbarger used Claude to coordinate contracts, logistics, and experiments with 11 external laboratories, alongside other operational tasks. In less than 18 months, Mithril moved from an idea for treating one form of the disease to encouraging results in mice. Anthropic says human trials for a drug candidate could begin as soon as late 2028 if results continue to be positive.

The company is also setting up ways for researchers to use its models beyond these projects. It launched Claude Science, an AI workspace for scientific research, offers Claude credits to life-sciences researchers, and created a Life Sciences Verification Program that lets verified organizations use its most capable models with more permissive safeguards. Anthropic has also opened applications for invited life-sciences researchers.

What the early results can—and cannot—show

I think the strongest evidence here is not that Claude has produced a drug, but that it can help with several kinds of work that often consume time: proposing candidate designs, reviewing records, and comparing trial plans. The Enveda result is especially concrete as a measure of labor saved. But catching more flagged issues in a document review does not establish that the entire regulatory process is faster, just as better laboratory measurements do not establish that a candidate will work in patients.

Anthropic says most of the programs are still preclinical. The announcement does not report human-trial outcomes, nor does it quantify how much Claude changed the total cost or duration of any drug program. Those are the measures that would show whether gains in individual tasks add up across development.

The company’s next test is whether it can move from assisting with discrete steps to supporting programs through clinical development. Anthropic says it plans to expand Claude’s role as projects advance and to work with organizations whose drugs have already reached that stage. Until then, its case rests on promising early results—and on whether those results survive the much harder transition from laboratory work to medicine.

Source

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X