Anthropic puts Claude at the service of a rare biotech
Anthropic acquired Mithril Medicine, a one-person biotech that pushed an experimental treatment for a rare childhood myopathy to mouse trial results. The case illustrates Claude's use in the lab but also its scientific limits.
Anthropic acquired Mithril Medicine, a biotech founded by Cody Tranbarger, after the latter developed an experimental treatment for a rare childhood myopathy in less than 18 months. The American company highlights one specific point: Claude was used to coordinate research spread across 11 laboratories, without ever replacing human decisions. The case is of interest to both health and AI sectors, as it shows how a model can accelerate scientific logistics without substituting experimental validation. And, in such stories, the lab mouse often ends up stealing the spotlight from the robot.
An acquisition that extends Anthropic's health strategy
Mithril Medicine has not only become a subsidiary of Anthropic; it also serves as a showcase for the use of Claude in drug discovery. According to information published by the company, the acquisition is part of a broader movement: in recent months, Anthropic's scientists have formed partnerships with six organizations developing new drugs, including Bristol Myers Squibb, Generate Biomedicines, and Enveda. Mithril was not part of this group of partners. Anthropic acquired it to conduct its own research programs.
The founder, Cody Tranbarger, confirmed on X the identity behind his pseudonym "Professor Oak" and summarized the deal in a very direct statement: "Surprise #1: Anthropic acquired a one-person biotech. Surprise #2: that person is me." The core of the matter remains scientific: the company says it obtained promising results in mice on a drug candidate intended to treat two rare forms of muscular dystrophy in less than 18 months.
Claude, conductor more than lone researcher
The 38-page technical report clearly nuances the image of a biotech run solo by an AI. Claude indeed helped read files, synthesize data, and prepare messages, but the human team retained control over decisions. According to Anthropic, the model analyzed a 341-page file in 7 minutes and 53 seconds, then proposed possible next steps for 113 candidate antibodies. Over the week, it drafted 17 technical emails; twelve were sent after review and correction by the team, and none left the system without human validation.
The process is more industrial than magical. Claude also suggested testing 24 antibodies outside of cell assays, for manufacturing reasons, before the team decided otherwise. In return, the model reduced its plan from 113 to 89 molecules and produced shorter email drafts. The main gain is therefore in coordination time: a first response arrived in 96 seconds, compared to 8.5 hours for the first feedback from an external lab, a delay that included nights, weekends, and 12 hours of time difference.
Useful but still fragile results
The picture remains far from definitive proof. The study has not been peer-reviewed, the mice were neither randomized nor observed blindly, and the main result is based on 7 treated animals versus 4 controls. The treated group was also tested last, after the animal facility moved, which may confound measurements. Another limitation noted in the report: treated mice had longer legs, a detail that may skew strength assessment.
The document also acknowledges a possible bias in the section describing the AI's work, as it was written by Claude under the team's direction. In short, the performance narrative is produced by the organization that benefits from it. Anthropic speaks of a "pipeline-in-a-person" model, referring to a portfolio of drug candidates managed by a single person, supported by multiple instances of Claude. The idea is appealing to the pharmaceutical industry, but it will still need to be tested outside the lab and beyond the press release.
What the pharmaceutical sector is watching closely
The Mithril Medicine case matters less for its immediate promises than for the method. It shows that AI can reduce coordination delays between labs, speed up technical writing, and help sort working hypotheses. However, it validates nothing alone, performs no physical tests, and bears no scientific responsibility. This clear boundary in the report matters as much to regulators as to investors.
Anthropic now plans to continue this research and launch several new programs. The company says a drug could enter human clinical trials by the end of 2028 if candidates continue to show good results. [to be verified] At this stage, the real question is not whether AI replaces researchers; it is rather to measure how far it can accelerate discovery without derailing the method.
By the numbers
- 11 laboratories — coordination announced by Claude, according to Anthropic.
- 38 pages — length of the technical report cited by the company.
- 7 treated mice and 4 controls — main result of the animal test.
- 18 months — claimed time between idea and results in mice.
- 2028 — projected horizon for possible human clinical trials.