Anthropic acquires first in-house drug programme from 'pipeline-in-a-person' company

Anthropic's first drug development programme is, fittingly, one that received help from AI from the start. But it did not use the drug-discovery algorithms that have caught the attention of large pharmaceutical companies. Instead, one person used Anthropic's AI assistant, Claude, to help with the early work of developing bispecific antibodies for two rare muscle diseases. Claude helped manage the practical challenges of developing drugs in a research model called "pipeline-in-a-person."

The AI company announced on Tuesday that it had acquired Mithril Medicine, taking over its muscle disease programme and hiring its founder and only employee, Cody Tranbarger.

Tranbarger used Claude to organise contracts and coordinate experiments carried out by 11 outside laboratories. He relied on the AI to help move information, including experiment materials, documents and data, between all the groups. His goal was to develop treatments for LAMA2-related congenital muscular dystrophy (LAMA2-CMD) and α-dystroglycanopathy (αDG).

By using Claude to manage the practical work, Mithril developed early evidence that a treatment based on bispecific antibodies could work against LAMA2-CMD in less than 18 months. More than 100 possible treatment candidates are now being studied in the lab using living cells or other biological material (in vitro).

Evidence supporting bispecific antibodies

Both LAMA2-CMD and αDG involve a problem with the connection between muscle fibres and the protective layer of proteins that holds them in place and helps prevent them from tearing.

Mithril's experimental bispecific antibodies are designed to attach to two things at once. One end attaches to dystroglycan in the outer membrane of a muscle cell, while the other attaches to laminin in the protective protein layer. Together, they form a bridge that holds the muscle fibre firmly to that layer.

Results from an early experimental molecule suggest that Mithril's approach may work. In a mouse model of LAMA2-CMD, mice given the highest dose of the experimental bispecific antibody had about twice the grip strength and muscle force of untreated mice. The treated mice also lived longer. Four of the eight mice in the comparison group died between two and six weeks of age, while only one of the eight mice given the experimental treatment died.

Now that Mithril's muscle disease programme is part of Anthropic, the company expects to name possible drug candidates for LAMA2-CMD and αDG next year. It plans to begin clinical trials for LAMA2-CMD by late 2028 and for αDG as early as 2029.

Testing the 'pipeline-in-a-person' model

Alongside its work on muscle diseases, Anthropic plans to use Mithril to test whether the pipeline-in-a-person model can help one scientist work on several biological ideas at the same time with Claude's support.

Anthropic's vision is for the scientist to choose the scientific direction and make all the key decisions, including which experiments to run and whether a drug programme should continue. Claude would keep track of each programme and coordinate with outside laboratories to keep the work moving.

The company said Tranbarger will also take on several more programmes focused on rare diseases in children. This will help test whether one person can manage an entire drug development pipeline.

In a post on X about Mithril's acquisition, Tranbarger said he "unequivocally" believes that combining AI and biology will cure diseases. But he believes this will happen through "orchestration and parallelisation"—organising work and running several programmes at the same time—rather than simply making development timelines shorter.

"Give [drug hunters] the bandwidth to run several programmes at once, and the time to focus on scientific judgment, and I'm willing to bet they'll make better calls," he wrote. "And at scale, better calls translate to more drugs, higher hit rates, broader patient impact, and likely, a disease or 10 or 100 cured."