Anthropic bought a one-person biotech that used Claude to get a muscular dystrophy drug to mouse results in 18 months

Anthropic said Tuesday, Oct. 27, 2026, that it acquired Mithril Medicine, a one-person company developing treatments for rare childhood muscular dystrophies, and published early results from six drug-development partnerships that use Claude.

The headline isn't that a chatbot designed a drug; it's that one scientist with Claude ran what normally takes a small biotech's worth of coordinators, contracts and lab wrangling, and got real mouse data in under 18 months. If that holds, the cost of starting a rare-disease program collapses, which is exactly where the money has never wanted to go. The skeptic's note writes itself: mice are not children, every number here is self-reported, and the earliest human trial is two years away. But Anthropic buying the company instead of just bragging about a customer says it wants to own the proof — and OpenAI's Orchestra Bio deal last week says the paperwork-and-pipeline war in drug development is now a two-lab race.

Anthropic said on Oct. 27, 2026, that over the past several months its scientists have run deep research partnerships with six organizations developing new medicines: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines, and The Column Group. It also started drug discovery of its own. Those partnerships, and that in-house start, are Anthropic's, from its research post the same day.

Anthropic acquired Mithril Medicine in July and disclosed the deal on Oct. 27. Before the deal Mithril had one full-time employee, founder Cody Tranbarger. He used Claude to organize contracts, logistics, and experiments with 11 external labs, and he delegated much of the company's operations to Claude. Those lines are Anthropic's.

Mithril's first program is a possible treatment for LAMA2-related congenital muscular dystrophy, shortened to LAMA2-CMD. Congenital means the disease shows up in early childhood. It is a rare muscle disease. The candidate is a bispecific "bridging" antibody. Bispecific means one antibody built to grab two targets at once. Mithril took it from idea to promising mouse results in under 18 months. In the chart Anthropic published, treated mice were about twice as strong as untreated ones on grip strength and on triceps muscle force. Those results are Anthropic's, and they are in mice.

Mithril's second program targets α-dystroglycanopathy, another rare muscle disease. The plan is to name formal drug candidates for both diseases in 2027. A drug candidate is the molecule a company picks to develop. It is not yet a medicine a doctor can prescribe. If those candidates hold up, human trials in LAMA2-CMD could start as soon as late 2028, and trials in α-dystroglycanopathy as soon as 2029. Tranbarger now continues the work at Anthropic and plans several more programs for rare pediatric diseases with no approved therapy. Pediatric means the patients are children. Those timelines are Anthropic's.

On manufacturing paperwork, Anthropic's post reports an Enveda test. Over 1,061 pages of manufacturing documentation, Claude caught all 11 issues human reviewers had found and flagged seven they missed. Human review time fell by 89 percent, from over 25 hours to under three. Claude also wrote the first draft of one regulatory document in four hours. By hand, that draft took about three months. A regulatory document is the paperwork a company files so a reviewer can check how a drug is made. Those figures are from Anthropic's post. Enveda's release the same day is the partner's own account of the collaboration.

BigHat Biosciences compared Claude-assisted antibody versions with its original leads. The best new versions bound more than 10 times more tightly. Binding is how hard the antibody holds onto its target. One stayed in solution at a temperature 18°C higher, meaning it remained dissolved when the older version would have fallen out. One clumped more than 100 times more slowly. Clumped means stuck to itself. Those comparisons are in Anthropic's post.

At Generate Biomedicines, scientists used Claude Code to propose a CAR-T design element about 70 percent shorter than the standard version. CAR-T is a cancer treatment that retunes a patient's own immune cells so they hunt tumor cells. In lab tests the shorter piece performed better: roughly 10 times as many receptors on the cells, and more efficient killing of tumor cells. A receptor is the sensor on the cell that recognizes the tumor. Those results are in Anthropic's post. They are lab tests, not patients.

The Column Group is creating a new company, Pleiades, that will use Claude to run drug programs more complex than a small team usually could. Bristol Myers Squibb is using Claude to find bottlenecks in clinical trial planning. A bottleneck is the step that slows the whole plan down. A clinical trial is the test in people. Those lines are Anthropic's.

Anthropic says the work is early and mostly preclinical. Preclinical means the tests are still in labs and in animals. Nothing in this set of results has been tested in humans yet. That limit is Anthropic's.

The picture is Anthropic's chart of the LAMA2-CMD mouse results. Grip strength, in grams of pull per gram of body weight, reads 0.70 for untreated mice (4 mice), 1.35 for treated mice (7 mice), and 1.82 for healthy mice (12 mice), with a bracket marked about 2×. Triceps muscle force, in millinewtons, ends at 24 for untreated mice, 47 for treated mice, and 82 for healthy mice, also marked about 2×. The same chart counts survival to six weeks of age: 4 of 7 untreated, 7 of 8 treated, and 12 of 12 healthy. The yellow caption reads "Treated mice were about twice as strong." The credit is Anthropic.

In plain terms, Anthropic said on Tuesday that it bought a one-person biotech in July, and that the founder, with Claude and 11 outside labs, got a rare-disease antibody to mouse data in under 18 months. The chart says the treated mice were about twice as strong. Six other partnerships reported early lab and paperwork gains. The earliest human trial named in the post is late 2028, and only if the candidates hold up. No person has taken any of this yet.