Anthropic puts Claude to work on drug development: here are its first results
Anthropic spent months working with six organizations developing drugs for cancer, food allergies and rare diseases. The first results are promising, although almost all of them are still at the lab stage.

A one-person company took a potential treatment for a rare form of muscular dystrophy from idea to promising results in mice in less than 18 months. Its founder, Cody Tranbarger, coordinated contracts, logistics and experiments with 11 outside labs with help from Claude, Anthropic's artificial intelligence assistant. The company is called Mithril Medicine, and Anthropic has just acquired it. It is one of the cases the company presented in a report on its effort to speed up drug development with AI.
According to Anthropic, over the past few months its life sciences team worked with six organizations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. The underlying idea is that creating a drug is not just about designing a new molecule, the task most associated with AI. You also have to choose what to target in the body, tweak the molecule so it can be manufactured at scale, prepare the documents regulators require and plan the clinical trials. In Anthropic's view, the bottlenecks are spread across that entire path, and the best way to find them is to get involved in real projects.
From cancer to allergies, case by case
Generate Biomedicines is looking to expand the use of CAR T therapy, a treatment that modifies a patient's immune cells so they attack cancer. So far it has mainly worked against blood cancers, and the company wants to bring it to solid tumors. To do that, it needs to shorten part of the therapy's genetic instructions to free up space for others. Six of its scientists used Claude Code, Anthropic's coding tool, to propose new designs. One turned out to be about 70% shorter than the standard version. In lab tests, it allowed immune cells to produce about 10 times more receptors (the proteins they use to detect cancer cells), killed tumor cells more effectively and kept the immune defenses active for longer.
Divergence Labs is working on a new approach to food allergies. These are triggered by a molecule called IgE, some of which is attached to our immune cells. Standard drugs neutralize the IgE circulating in the blood. Divergence wants to detach it from the cells entirely, which requires designing an antibody with two "arms." Claude turned those goals into a design plan and, largely autonomously, generated about 35,000 designs in roughly a month. Early tests turned up three designs capable of binding to IgE and 20 capable of dislodging it. Three of the latter appeared to do so better than the standard drug. The next step is to combine both arms into a single drug.
At BigHat Biosciences, Claude was connected to the data system of the company's automated lab and proposed changes to two candidate molecules. In about 10 weeks, roughly 1,010 variants were made and tested. Midway through the process, the team suspected that the usual stability measurements were misleading. Claude and a researcher reanalyzed the data and proposed other metrics, and the main predictions were confirmed. The best versions bound to their target more than 10 times more strongly than the originals.
Enveda's case is less flashy but very concrete: paperwork. Its manufacturing chemists spent up to 75% of their time on documentation. In tests on 1,061 pages, Claude detected the 11 problems human reviewers had found, plus seven more they had missed. Human review time dropped 89%, from more than 25 hours to less than three. It also wrote, in four hours, the first draft of a regulatory document that took about three months to produce by hand.
What still needs to be proven
Anthropic is also working with The Column Group, a biotech investment firm, on Pleiades, a new company that will use Claude to run more drug programs than is typical for a small team. With Bristol Myers Squibb, it is evaluating early-stage clinical trial plans and plans to expand that work in 2027. Michael Pourdehnad, a vice president at the drugmaker, said that with a single prompt Claude generated an initial analysis that would otherwise have taken days.
These figures should be read with caution. The results come from Anthropic and its partners, and they rely on technical reports from the companies themselves, not on studies reviewed by independent scientists. The company also acknowledges that almost all of the work is preclinical, meaning it has not yet been tested in people. In Mithril's case, Anthropic estimates that a drug could begin human trials in late 2028, provided the candidates continue to show good results.
For researchers outside these companies, Anthropic points to Claude Science, an AI-powered workspace for researchers; usage credits for life scientists; and an open call for groups that want to use Claude on drug discovery questions. It also created a verification program that gives trusted organizations access to its most powerful models with more flexible safeguards, designed to prevent misuse without holding back legitimate research. The company did not specify requirements, amounts, or whether these options are open to labs and universities in Latin America. That will be the first thing worth checking to find out whether the region can take part.
Source: Anthropic