Anthropic Buys MITHRIL MEDICINE
In a Nutshell
Anthropic acquired Mythril Medicine, a one-person AI-powered biotech startup that advanced a rare muscular dystrophy treatment from concept to mouse results in under 18 months, while collaborating with six institutions to demonstrate AI accelerating multiple stages of drug development. The report argues that personalized medicine for rare diseases, cancer, and aging requires a million-fold increase in computing power within about 10 years to match the scale of 10,000+ rare diseases and the complexity of individualized treatments. This scale-up would enable AI to perform the work of expert teams for every patient, shifting the bottleneck from human expertise to computational resources needed for reading biology, designing therapies, and running experiments at population scale.
These notes were generated by AI and may contain inaccuracies.
On October 27, 2026, Anthropic announced the acquisition of Mythril Medicine, a company consisting of only one person. Mythril Medicine has been developing treatments for two rare types of muscular dystrophy. Its founder, Cody Tranbarger, used the Claude model to organize contracts, logistics, and experiments with 11 external laboratories. Claude was also entrusted with many other aspects related to the management of the company. In less than 18 months, Mythril took a potential treatment for one of those diseases from just an idea to promising results in mice. This demonstrates what one person can do using artificial intelligence today. Tranbarger now works at Anthropic and plans to start several new programs while there. Anthropic expects one of its drug candidates to be ready for human trials by late 2028. This story comes from a report published by Anthropic on the same day, titled "Early Progress in Accelerating Drug Development Using Cloud".
This report is considered the clearest picture yet of why a million times more computing power than currently available is needed. Over the past few months, the life sciences team at Anthropic has worked closely with six institutions developing new drugs. Anthropic's view is that there is no single problem to be solved in drug development. The obstacles are distributed throughout the entire process.
Five organizations each tackled a different obstacle in drug development:
There is a treatment called CAR-T cell therapy, in which doctors take immune cells from the patient himself, reprogram them to attack the cancer, and then return them to the patient. So far, this treatment has mostly helped people with blood cancers. Generate Biomedicines wants this treatment to work on solid tumors as well. To do this, Generate attempts to shorten one part of the genetic instructions for treatment. This will allow space to be made available for more instructions directed at other types of cancer. Six scientists at Generate used the Cloud code feature to propose new designs for this part. One of the designs was about 70% shorter than the standard version. In laboratory tests, it proved to work better. With this design, the immune cells produced about 10 times the number of receptors. This means that it killed tumor cells more efficiently, and remained active for a longer period.
Food allergies are caused by an antibody called IgE. This antibody binds to some of your immune cells, leading to an allergic reaction. Today's standard medicines target only the antibodies (IgE) that swim freely in your blood. Divergence Labs wants to completely remove these antibodies from immune cells. This requires a very precise design with two arms. One arm holds onto the IgE body, while the other arm removes it from the cell. In about a month, Claude designed 35,000 antibody designs. Divergence then narrowed down and tested those designs. In those initial tests, three designs were able to catch the IgE, while 20 succeeded in removing it. Three of those twenty appeared to be able to remove IgE better than standard drugs.
Big Hat Biosciences used Cloud to improve two antibody drug candidates. Their lab created and tested over a thousand versions of those candidates in about 10 weeks. The best versions were more than 10 times more likely to achieve their goals than the original versions.
To license a new drug, the company needs records proving that each batch was manufactured according to strict quality standards. The company also has to submit applications to regulatory authorities. At Nvidia, manufacturing chemists were spending up to 75% of their time on those papers. Therefore, Nvidia created a system that links all of its manufacturing documents to the cloud. Cloud then identifies any potential problems. In a test on 1,061 pages of documents, Cloud discovered all 11 problems that human reviewers found. Claude also identified seven additional problems that human reviewers had missed. Using Cloud, preparing a single regulatory document took only 4 hours. Manually, that document would have taken about 3 months.
Bristol-Myers Squibb is one of the world's largest pharmaceutical companies. The company uses Cloud to plan clinical trials and estimate their chances of success. A senior company executive said that, based on a single directive, Cloud provided a strong preliminary analysis that would otherwise have taken days.
All of this research is still in its early stages, and most of it has not yet reached human trials. Many drugs that look great in the lab never reach patients. However, the aim of this report is not a specific drug. The goal is how many ideas a small team can try right now and how quickly they can implement them.
There are very few patients who actually receive treatment specifically designed for them. In the summer of 2024, a baby boy named K.J. Mouldon was born near Philadelphia. When he was in his early days, his doctors discovered that he had a rare genetic disorder called CPS1 deficiency. His liver lacked an enzyme the body needed to convert ammonia into urea. Without this enzyme, ammonia accumulates in the blood, and at high levels, it can damage the brain and liver. In the worst case, such an accumulation can cost you your life.
Since 2023, a team at Children's Hospital of Philadelphia and Penn Medicine have been working together on gene editing for disorders such as K.J.'s condition. Within 6 months, this team designed and manufactured a gene-editing treatment for K.J. alone. This treatment was designed to enter K.J.'s liver cells and correct a single incorrect letter in his DNA. K.J. received the first dose in late February 2025. He then received a second dose in March and a third in April. So far, that treatment has been working successfully. He can now eat more protein and needs less medication. His doctors say he is growing well and thriving. After more than 300 days in the hospital, K.J. returned home. This is personalized medicine. It is a medicine made for one person only.
In February 2026, the Food and Drug Administration proposed a new method for approving treatments such as K.J.'s therapy.
In the United States alone, approximately 30 million people live with a rare disease, which is equivalent to one in ten Americans. Scientists know of more than 10,000 rare diseases, but only about 500 of them have a cure. We could never provide a team like K.J.'s for every single one of these patients. Simply put, there are not enough experts in the world. The bottleneck is not science itself. The bottleneck is the number of expert teams we have.
Once artificial intelligence can do most of the work of a team like K.J.'s, the problem will no longer be the number of experts we can train. The bottleneck is the amount of computing power we have. This is not limited to rare diseases. CAR-T cell therapy is truly personalized because doctors make it from the patient's own immune cells. Each cancer is different from the other. Each tumor carries its own set of mutations. Even the way your body deals with a certain medication is slightly different from how others do. Therefore, in the long run, every element deserves this level of customization. This includes cancer, heart disease, and the slow damage of aging itself.
One way to imagine computing power increasing a millionfold is to look at Divergence's numbers. In about one month, Claude created 35,000 antibody designs. Multiply this capacity by a million and you get 35 billion designs every month. There are approximately 8 billion people on Earth. This means more than four designs for every person on the planet every month.
The computing power used to train the largest artificial intelligence models has increased four to five times annually since 2010. This has been going on for 16 years. If computing power were to grow fourfold each year over the next ten years, that would result in a million-fold increase. So, a million is not science fiction. We've done it before, and we'll do it again. At the same rate, producing computing power a million times greater would take only about 10 years.
The concept of aging escape velocity was popularized by the gerontologist Aubrey de Grey. Currently, old age is still winning. With each passing year, you lose a year of your remaining average lifespan, and medicine only restores a small fraction of that year. The escape rate from aging is the point at which medicine gives us more than one year of average life expectancy for every calendar year that actually passes. At that point, you will no longer be in a race against time.
Over the past 160 years, the global average lifespan for women has increased by approximately 3 months each calendar year. This is one of the greatest achievements in human history. But the speed at which we escape old age requires an additional 12 months of life each calendar year. This is four times faster than what we are achieving now. Those gains must reach people of all ages living today.
In order to progress four times faster, we need to deal with what aging actually is. Aging is not just one disease. These are many types of damage that accumulate, and this damage increases at different rates for each of us for different reasons. To overcome this damage, you will have to measure each person's body and find out what defect they have. After that, you will have to fix those failures over and over again for the remainder of that person's life. This means providing personalized care for each individual throughout their life.
Each step in that process depends on computing power. It requires computing power to read your biology, design your treatment, and conduct experiments that prove the treatment is effective and safe. The speed at which you escape old age is related to speed. In medicine, the speed of research has begun to depend on the amount of computing power we have. When a million times more computing power is needed, that means everyone on Earth should have a team like K.J.'s. Science is already at work and regulators are making way for this new acceleration. What we do not yet have is sufficient computing power. Increasing computing power a million times is how we get to personalized medicine and speeding up the aging process.
Keep David Shapiro in your library
Save the videos and channels worth coming back to, and find them again in one place.
