Anthropic, the company behind the Claude AI models, says it is now running drug research of its own. In a post on Oct. 27, it said that in July it bought Mithril Medicine, a company with only one full-time employee, its founder Cody Tranbarger. Part-time consultants and outside labs did the rest of the work, and Anthropic says Tranbarger used Claude to organize contracts, logistics and experiments with 11 outside labs, and handed many other parts of running the company to Claude. Mithril's two programs, for two rare and severe forms of muscular dystrophy, are now Anthropic's first in-house drug programs.
Mithril's first disease is LAMA2-related congenital muscular dystrophy, which Anthropic says affects an estimated 4,000 to 5,000 people in the U.S. and Europe and has no approved treatment. Most patients are severely weak from birth, and few live to 30. Each muscle fiber is held in place by a sheet of protein that keeps it from tearing as the muscle works. In this disease the anchor between fiber and sheet is too weak, so fibers are easily damaged, and over time the muscle dies and is replaced by fat and scar tissue. Mithril's possible treatment is an antibody, a Y-shaped protein of the kind the immune system makes. Each one is designed to grab the muscle fiber with one arm and the sheet with the other, clamping the two back together.
Anthropic says Mithril took the idea to promising results in mice in less than 18 months. In a short study, mice with a severe form of the disease that got the highest dose had roughly twice the grip strength and muscle force of untreated mice at six weeks old, and more of them survived. The mice got early, rough versions of the antibody, made only to test whether the idea works at all, not the final drug. The treated mice were still weaker than healthy ones, so it was not a cure. Mithril aims to choose its drug candidates in 2027 and, if they hold up, to start trials in people as soon as late 2028 for the first disease and 2029 for the second, a related disease in which the muscle's side of the anchor fails. Anthropic says dates after 2026 are targets, not commitments.
Anthropic calls the setup "Pipeline-in-a-Person": one scientist tries to run several drug programs at once while Claude keeps the records, coordinates the outside labs and keeps the work moving, and the person decides which experiments to run and whether a program goes on. Anthropic calls Mithril the first test of whether this works. It says the approach could make it affordable to work on very rare diseases that a full research team would cost too much to pursue, and notes that Claude cannot make cells grow or mice age any faster.
The same post describes work with six partner organizations, including the drugmaker Bristol Myers Squibb. At Enveda, Claude checked 1,061 pages of drug manufacturing records. Across several runs it caught all 11 problems human reviewers had found, one run flagged seven more they had missed, and the time people spent reviewing fell by 89%, the companies say. At Generate Biomedicines, scientists using Claude designed one part of a cancer treatment made from patients' own immune cells. Generate says its version was about 70% shorter than the standard part, and worked better in lab tests. Anthropic gives these partners Claude credits and engineering help.
Anthropic says the work is still at an early stage and mostly has not reached tests in people. Results in mice do not show that a treatment is safe or works in patients, and the mouse study was small. According to a news report, the study has not yet been checked by outside scientists, and the treated and untreated mice were tested months apart, which makes it harder to credit the gains to the treatment; the researchers call for a repeat study. The reports describing the work are the companies' own, and Anthropic has not said what it paid for Mithril.