On 27 October 2026 Anthropic, the company that makes the AI model Claude, said it had bought Mithril Medicine, a drug start-up with just one full-time employee. The same day, its founder, Cody Tranbarger, revealed that he is the person behind "Professor Oak", an X account known for its biotech commentary, which he had kept anonymous for years. Last checked 30 October 2026.
Solid dots have happened. Hollow dots are Anthropic's own estimates.
Nathan Frey, who leads drug research at Anthropic, told Endpoints News that Anthropic first wanted the versions of Claude that Mithril used to be open to other approved science organisations, through a scheme it calls the Life Sciences Verification Program. It also wanted examples from other companies, to show this way of working isn't limited to a one-person start-up.
Anthropic says Mithril went from an idea to promising mouse results in under 18 months. That looks like a reasonable pace rather than a record. Mithril copied the two gripping ends of its antibody from antibodies already described in patents, so it skipped much of the early work that a 2010 industry model (Paul and colleagues) allows years for. Mithril hasn't published dates for its experiments, so the 18 months can't be checked. None of its work has been peer reviewed, meaning checked by independent scientists.
Cody Tranbarger
Runs "Professor Oak" on X
Roles before Mithril as listed on his LinkedIn profile.
Cody Tranbarger came to biotech through finance and business deals rather than the lab. In June 2024 he joined Atlas Venture, a biotech investor in Cambridge, Massachusetts, as an entrepreneur-in-residence: someone an investor backs to start a new company. He founded Mithril that December. Atlas put in a small first round of funding and is Mithril's only named investor, and Atlas partner Jason Rhodes chaired its board.
Nathan Frey, now head of drug research at Anthropic, met Tranbarger in the first week of his own start-up, Coefficient Bio, which Anthropic bought in April. Frey was a fan of Professor Oak but had no idea who ran it. He told Endpoints News he should have guessed: when they met, Tranbarger asked him to "choose your starter: small molecule, antibody, gene editor", three kinds of drug. In the original Pokémon games, Professor Oak asks each new player to pick a starter Pokémon. He found out later, when he mentioned one of Professor Oak's analyses and Tranbarger replied, "Oh yeah, that's me."
Tranbarger is now part of Frey's team, where he says he will lead Anthropic's wider in-house drug development.
Mithril's main target is LAMA2-related congenital muscular dystrophy, or LAMA2-CMD, a rare inherited muscle disease that starts at birth. Children with it can't make a working version of laminin-211, a protein that anchors each muscle fibre to a thin sheet wrapped around it. The body uses a weaker substitute, laminin-411, which holds on poorly. As the muscles work, the fibres pull loose and die.
42 children in one London study classed as making no laminin-211, from muscle samples or, without one, from their genes and symptoms. Each grid shows the same 42 children.
Source: London study cited in Mithril's technical report. Anthropic says few patients live to 30.
About 1 in 120,000 births
That is the worldwide estimate from a 2023 study, worked out from gene changes in large genetic databases rather than by counting patients. For people of non-Finnish European ancestry it is about 1 in 100,000. The authors call it a minimum, because it leaves out the gene changes thought to cause milder forms.None approved
Only one experimental drug, omigapil, has been tested in patients, and only for safety. Anthropic says its development stopped in 2021.No results yet
Mithril also aims at a related group of muscle diseases, the alpha-dystroglycanopathies. A similar antibody made by another research group helped mice with one of them.Instead of replacing the missing protein, Mithril designed an antibody that straps the weak substitute to the fibre. Antibodies are the Y-shaped molecules the immune system makes, and drug makers can engineer them to grab chosen targets. One end of Mithril's antibody holds the fibre's anchor point, beta-dystroglycan (β-dystroglycan). The other end holds laminin-411.
A simplified picture, not to scale. Some children with the disease also have nerve and brain problems, which the antibody would not treat.
After a lab-dish test showed the antibody could make cells stick to the gripping end of laminin-411, Mithril ran one small study in mice bred to have the disease. Mice given the better of two antibody designs ended up with about twice the muscle force and grip of untreated mice. The study didn't examine muscle tissue or measure how much of that design reached the muscle, so it can't show the bridge forming there.
Day each group of mice was tested, counted from the first test of untreated sick mice. Fair studies test groups side by side, because readings drift over time.
Source: Mithril Medicine technical report, supplementary table S6
Front-leg muscle force at about six weeks old, in millinewtons. Groups were tested on different days. Bars show the average; thin lines show each group's spread.
Grip strength at six weeks, in grams. Groups were tested on different days, and healthy mice's grip itself varied between sessions.
The other antibody design (5 mice) gave about the same muscle force as no antibody (1.08 times the untreated average), though its grip was higher (11.8 g against 6.7 g).
What a convincing mouse study would have, against what this one had
Mithril's team says the next step is to repeat the study blind, with treated and untreated mice tested side by side.
Mithril had one full-time employee, but it wasn't one person working alone. It had money and a board from Atlas Venture, eight part-time consultants and advisers, and hired labs that did every experiment (11 of them, Anthropic says). Claude read the lab data and emails, ran the analyses and drafted replies. A person decided what was sent.
Mithril picked its best recent week as an example. In it, Claude:
The 113 new designs, as Claude sorted them
By the end of the week, the cell tests on the new designs hadn't started, and the mouse studies were still at the draft-email stage. It was a week of organising.
Mithril examined six of Claude's mistakes from that week. Claude found five of them itself, one only after a lab asked a question. At least three still went out in emails a person had edited and sent.
The two slow ones were wrong statements in Claude's review of a lab's data. Source: Mithril Medicine technical report
Time to turn a first idea for one new project into a full plan. The darker part of each bar is the low end of the range.
One example. The figure without Claude is Tranbarger's own estimate.
Mithril is tiny, but it gives Anthropic drug projects of its own.
Drug projects one person can run with Claude, by the informal reckoning of Nathan Frey, Anthropic's head of drug research. In the second row, solid squares are the low end of his estimate and dashed ones the upper end.
Frey says Claude can't yet run a project with nobody in charge. Anthropic compares what Claude would do next with what the expert running the project does, and the expert's skill is still essential. Source: Endpoints News, 30 October 2026
Anthropic makes its money selling access to Claude. Now it owns drug projects too, and Frey says Mithril's are its only ones in animal studies. Anthropic filed to list its shares on the stock market the day before the news. The filing doesn't mention Mithril, but warns that possible drugs may never make money and that AI mistakes in healthcare could harm patients.
The drug-industry news site Endpoints News reports that the share sale is expected to be the largest ever. In Forbes, Jon Markman argued that investors worried about Anthropic's slowing revenue growth "may be overlooking a bigger question": how much Claude could be worth if it can help run complicated work in other industries. That is one writer's opinion. Nothing on this page is investment advice.
What was new here was the organising. If one person working with Claude and hired labs can run a drug project, small teams may be able to start early projects faster and more cheaply. People still did the lab work and made the decisions, though.
No patient has received the antibody, and most drugs that reach tests in people are never approved. Anthropic's life-sciences leaders had told Endpoints News they don't plan to run such trials themselves. In an interview published on 30 October, Frey said it was too early to know: Anthropic is "certainly capable" of running them, and is also open to handing projects to a partner.
Anthropic's filing pitches an "age of miracles", Endpoints News says, with possible cures for cancer, Alzheimer's and heart disease. Asked if that promises too much, Frey said he couldn't tell how long such treatments would take without AI, only that AI could bring them sooner. He'd rather count the years of healthy life patients gain than argue over what counts as a cure.
A week earlier, OpenAI, the maker of ChatGPT, said it had bought Orchestra Bio, whose software helps drug developers coordinate their work, much like Claude's job at Mithril. In April Anthropic bought Coefficient Bio, and Vas Narasimhan, chief executive of the drug maker Novartis, joined its board. Frey says Anthropic has no other purchases to talk about for now.
Claude's mistakes got past the person checking its emails. And almost everything known about Mithril comes from Anthropic, which owns it, or from Mithril's own report, which was written with Claude's help. None of it has been peer reviewed, and no one outside the company has repeated the mouse study.
The usual stages of drug development, and where Mithril's antibody stands. The dates are Anthropic's own estimates, and they could slip.
Sources: Anthropic; Endpoints News; BIO, Informa Pharma Intelligence and QLS Advisors, Clinical Development Success Rates 2011-2020
Leave your email and I’ll send you the next explainer when it’s ready, with every chart and its source.
You’ll get new explainers and, now and then, news of things I’m working on. Unsubscribe any time. Privacy notice