Claude in the lab: Anthropic shows early drug-development results and buys a one-person biotech

On October 27, Anthropic's life sciences team published an update on its company blog about how Claude is being used in drug development. The starting premise is stated plainly: according to Anthropic, there is no single problem to solve, because the bottlenecks are scattered across the whole process. Hence the decision to work with six partners (BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group) on very different stages, from molecule design to regulatory documentation to clinical trial planning, while launching programs of its own in parallel.

On the design side, the data Anthropic has released is striking, but all of it comes from the lab. At Generate Biomedicines, six scientists used Claude Code to shorten a genetic component of CAR-T therapies, with the aim of extending them to solid tumors: a design roughly 70% shorter than the standard let immune cells produce about 10 times more receptors, killed tumor cells more efficiently and stayed active longer. At Divergence Labs, working on food allergies, Claude generated around 35,000 antibody designs in about a month; in early tests 3 bound IgE and 20 managed to pull it off immune cells, and of those, three seemed to work better than the standard drug. At BigHat, hooked up to the lab's data system, Claude proposed changes to two antibody fragments: about 1,010 variants were produced and tested in roughly 10 weeks, with the best showing more than 10 times greater affinity for the target. When standard stability metrics turned out to be misleading, Claude and an Anthropic researcher proposed new metrics and a protocol change, a sign that the model is shaping the method too, not just the molecules.

The least glamorous part may be the most concrete. At Enveda, Claude reviewed 1,061 pages of manufacturing documentation, catching all 11 issues flagged by human reviewers plus 7 they had missed; human review time fell by 89%, from more than 25 hours to under 3 according to Anthropic (The Next Web says nearly 26). One of the two regulatory documents drafted by the model took 4 hours instead of roughly three months of manual work, while review of the other turned up a minor error. Enveda's founder and CEO, Viswa Colluru, said AI can help clear bottlenecks further downstream in development as well. Bristol Myers Squibb, meanwhile, uses Claude to assess plans for an early-stage drug and estimate the likelihood of clinical trial success, and intends to expand the work in 2027. The Column Group is building Pleiades, a company that plans to have Claude run more, and more complex, programs than a small team could handle; its first skill, designed to judge whether a therapeutic target is worth pursuing, was tested on 45 target-disease pairs.

Then there is the acquisition of Mithril Medicine, a biotech made up of its founder alone, Cody Tranbarger. According to Anthropic, Tranbarger used Claude to coordinate contracts, logistics and experiments with 11 outside labs, taking a candidate for a rare form of muscular dystrophy from idea to promising results in mice in under 18 months. According to The Next Web, the main target is LAMA2-related congenital muscular dystrophy; in mice, one antibody design roughly doubled grip strength compared with untreated animals. The terms of the deal were not disclosed, and Tranbarger has joined Anthropic. If the candidates keep performing well, Anthropic writes, a drug could reach human trials as early as late 2028.

The limits deserve to be stated as clearly as the numbers. Almost all the data comes from Anthropic or its partners, not from independent checks, and the technical report has not been peer reviewed; the molecular results are preclinical, obtained in the lab or in animal models, and the 2028 date is explicitly conditional. Anthropic itself writes that all of this research is still at an early stage. The Next Web also points out that the announcement came the day after the filing of the prospectus for a Nasdaq listing, in which drug discovery reportedly appears among the potential benefits. As next steps, the company lists expansion into clinical trials, Claude Science, credits for researchers in the field, a Life Sciences Verification Program and a call for visiting researchers.

What strikes me is not any single antibody but the geography of the work: a model that writes regulatory documents, double-checks manufacturing pages and coordinates eleven labs on behalf of one person. If the promise holds, the advantage will lie less in inventing molecules than in reducing the friction between one step and the next. To find out, though, we need the things this announcement still lacks: external verification, peer review and, sooner or later, patients. — Olya