Anthropic has acquired the small biotechnology company Mithril Medicine. Small is meant literally, as the company had just one employee, its founder Cody Tranbarger. He was developing treatments for two rare forms of muscular dystrophy while entrusting most of the company’s operations to artificial intelligence. Anthropic did not disclose how much it paid for the company. Along with the acquisition announcement, however, it published the first results of collaborations with six pharmaceutical and biotechnology companies and confirmed that it is embarking on the development of its own drugs.
Tranbarger used Claude to manage contracts, logistics, and experiments with eleven external laboratories. He also handed over to artificial intelligence a range of other tasks that would occupy entire departments at a conventional company. The result has been unexpectedly fast by pharmaceutical research standards. Mithril progressed from the initial idea to promising results in mice in less than a year and a half.
The company focused primarily on LAMA2-related congenital muscular dystrophy, a rare disease affecting children. According to a technical report published by Anthropic that has not yet undergone peer review, mice treated with one of the proposed antibody types had roughly twice the grip strength of untreated mice.
What Anthropic Plans to Do with Mithril
Tranbarger has joined Anthropic and is continuing his research. He is also preparing several new programs. If his drug candidates also perform well in further tests, the first of them could enter human clinical trials in late 2028. Claude is carrying out a large share of both the scientific and operational work in these programs.
Why is a company developing language models moving into pharmaceuticals? Anthropic explains that drug development does not have one specific bottleneck that can simply be accelerated. Drug development is not held back by a single problem; it is delayed at every stage, the company writes. Finding a suitable target in the body, designing a molecule, manufacturing it at scale, handling regulatory paperwork, and planning clinical trials all take years. According to Anthropic, the best way to identify these obstacles is by participating in real programs across different areas of medicine.
The company expects three things from this. It wants to gain a better understanding of where its models are strong and where they fall short, and it has already adjusted its future development priorities accordingly. The results are intended to serve as a demonstration for other biotechnology companies. If the experiments continue to progress well, they could lead to actual drugs for food allergies, rare childhood diseases, and several types of cancer.
A New Opportunity for Cancer Treatment
One of the partners is Generate Biomedicines, which is working on so-called CAR T-cell therapy. So far, it has primarily helped combat blood cancers, and Generate wants to modify it so that it also works against solid tumors. To do so, it needs to shorten part of the therapy’s genetic instructions to make room for additional ones.
Six Generate scientists used Claude Code to design new versions. One of them was about 70 percent shorter than the standard variant and performed better in the laboratory. It enabled immune cells to produce roughly ten times more of the receptors they use to recognize cancer cells, destroy tumor cells more effectively, and remain active longer.
Thousands of Allergy Antibodies in a Month
Divergence Labs is searching for a new way to treat food allergies. These are triggered by the IgE molecule, some of which circulates freely in the blood while some sits directly on immune cells. Conventional drugs neutralize only the free-floating IgE. Divergence wants to remove IgE from immune cells completely, which requires a drug with two arms. One must bind tightly to IgE, while the other helps pry it off the cell.
Claude converted the company’s goals into a detailed plan and then generated around 35,000 antibody designs more or less independently, in approximately one month. In the initial tests, scientists found three designs capable of binding IgE and twenty that released it from cells. Three of them apparently performed better than the standard drug. The company must now combine both arms into a single treatment.
One Thousand Variants in Ten Weeks
BigHat Biosciences combines artificial intelligence with an automated laboratory to improve promising compounds. These must bind tightly to their target, withstand manufacturing and storage, and resist aggregation. Connected to the company’s laboratory data, Claude proposed modifications to two antibody fragments, one of which targets cancer cells while the other is intended to help keep the drug in the body longer. Based on its designs, the laboratory produced and tested roughly one thousand variants in ten weeks.
Halfway through the work, the team began to suspect that conventional measures of thermal stability and aggregation were distorting the results. Together with an Anthropic researcher, Claude reexamined the raw data and proposed a different measurement method. The main predictions were confirmed. The best versions bound to the target more than ten times as tightly as the original, the anti-cancer compound remained dissolved even at a temperature 18 degrees higher, and the other aggregated more than one hundred times more slowly.
Less Grueling Paperwork
At Enveda, chemists responsible for manufacturing spent up to 75 percent of their time on paperwork. Before clinical trials, the company must document that every batch of a drug was produced according to strict quality standards and submit an application to regulators. Enveda therefore connected all its manufacturing documents to Claude and had it search for errors.
In a test involving more than one thousand pages of documentation, Claude identified all eleven problems found by human reviewers, as well as seven additional ones they had missed. A review that previously took more than 25 hours was reduced to less than three. Claude also wrote the first drafts of two important regulatory documents. It completed one in an astonishing four hours, whereas people would have spent roughly three months preparing it manually. Our collaboration with Anthropic shows that AI can remove bottlenecks even in the later stages of development, said Enveda founder and CEO Viswa Colluru.
From the Laboratory to Clinical Trials
More recently, pharmaceutical giant Bristol Myers Squibb also joined the partners. With Claude’s help, it assesses plans for early-stage drug clinical trials and estimates their chances of success. From a single prompt, Claude prepared a solid initial analysis that helped us evaluate various development options, a process that would otherwise have taken days, said Michael Pourdehnad, who leads early clinical development in hematology, oncology, and cell therapy at the company. The two companies plan to expand their collaboration over the next year.
According to Anthropic, everything is still only at the beginning. Most of the research is in the preclinical stage, meaning that, with a few exceptions, the drugs are not yet being tested in humans. The company wants to gradually extend Claude’s role into clinical trials and is seeking additional partners whose programs have already reached this stage. It offers scientists the Claude Science work environment, credits for researchers in biology and medicine, and access for vetted organizations to its most powerful models with more flexible safety safeguards, which can otherwise hinder legitimate biological research.
Sources: anthropic.com and theinformation.com
