Anthropic Acquires One-Person Pharma Company Mithril Medicine, AI Giants Rush into the Pharmaceutical Industry

A biotechnology company with only one full-time employee has been acquired by the AI large-model company Anthropic.

The company is called Mithril Medicine. Its founder, Cody Tranbarger, used Claude to coordinate research across 11 external laboratories, advancing a candidate therapy for a rare muscular dystrophy from concept to mouse experiments in less than 18 months, with preliminary positive results. These projects are now part of Anthropic's first in-house drug discovery programs.

On October 27 local time, Anthropic disclosed the acquisition and also disclosed a series of advances in AI-enabled drug discovery. Over the past few months, the company has collaborated with six drug discovery organizations, including Bristol Myers Squibb and Generate Biomedicines, attempting to have Claude participate in molecular design, experimental optimization, drug manufacturing and clinical trial planning. At the same time, Anthropic has also begun advancing drug discovery projects itself.

"New, heavyweight players are joining, and industry competition has become more intense," a relevant person from the AI pharmaceutical company XtalPi told a Yicai reporter, especially after coding became an important commercialization scenario for large models, leading AI companies need to find the next application field where model capabilities can be brought into play and commercial value can be created. Judging from this disclosure, Anthropic's cooperation with relevant organizations has long been very deep.

From providing AI tools to personally participating in drug discovery, Anthropic's business boundaries are changing. Technology companies such as OpenAI, Google and ByteDance are also entering this field. Why are large-model companies collectively optimistic about the pharmaceutical industry? How will they change this industry that has long faced high costs and high failure rates?

From assisting pharmaceutical companies to developing new drugs themselves

The research results Anthropic announced today cover multiple stages of drug discovery.

In the field of cell therapy, Generate Biomedicines used Claude Code to optimize CAR-T-related gene design. In one experiment, the relevant genetic instruction fragment was shortened by about 70%, and cell surface receptor expression increased by about 10 times.

In antibody development, Divergence Labs used Claude to generate 35,000 antibody candidate designs in about one month; BigHat Biosciences designed and tested 1,010 molecular variants in about 10 weeks, and some optimized antibodies showed several-fold improvements in target binding ability.

In addition to molecular design, Anthropic is also trying to bring AI into the daily processes of drug discovery. Partner Enveda used Claude to review 1,061 pages of drug manufacturing documents, identifying all 11 issues found by manual review and additionally flagging 7 potential issues; manual review time was shortened from more than 25 hours to less than 3 hours. Bristol Myers Squibb used Claude to analyze early clinical development plans and assist in evaluating different trial paths.

But these cases reflect one of Anthropic's judgments: the bottleneck in drug discovery does not exist only at the molecular discovery stage, but is also dispersed across experimental design, data analysis, manufacturing, regulatory filings and clinical trials. If AI can participate in the entire process, the potential value is not just helping scientists find more candidate molecules.

This also explains why Anthropic began developing drugs itself.

The acquired Mithril Medicine mainly studies treatments for two rare muscular dystrophies. The founder used Claude to manage contracts, logistics and experimental arrangements for external laboratories, enabling a very small team to organize complex drug research. Anthropic estimates that if subsequent research progresses smoothly, one of the candidate pipelines could potentially begin human clinical trials as early as the end of 2028.

Previously, Anthropic had already acquired Coefficient Bio and established its own biological laboratory. Compared with only providing model services to pharmaceutical companies, direct participation in research and development allows the company to obtain real experimental feedback to test Claude's capabilities and improve the model.

Related arrangements also appear in Anthropic's prospectus submitted this week. Anthropic lists life sciences as an important application market for frontier AI. As of September 30, nine of the world's top ten pharmaceutical companies were Anthropic customers, and each had paid Anthropic at least $100,000 in the past 12 months.

Anthropic also stated in the prospectus that it is developing AI agents to support scientists in completing work at different stages of drug discovery. However, the company also warned that the life sciences business may bring risks such as laboratory safety, regulatory compliance and R&D failure. It has not yet disclosed the independent revenue and investment scale of this business.

For a company that mainly relies on model services for revenue, developing drugs itself means a longer investment cycle and more uncertain returns.

Anthropic is not the only foundation model company entering the pharmaceutical industry.

On October 20, OpenAI announced the acquisition of the drug discovery collaboration platform Orchestra Bio and the integration of its technology with the life sciences models GPT-Rosalind and Rosalind Workbench.

Orchestra Bio's AI agents are mainly used to coordinate complex drug R&D projects, such as tracking experiment progress, analyzing the impact of manufacturing delays on clinical trials, and helping R&D teams adjust plans. OpenAI hopes to let AI not only participate in scientific discovery, but also manage a large amount of professional work between experiments and clinical development.

English text is being prepared.

Chinese internet giants are also making moves. Earlier reports said that Anew Labs, an AI pharmaceutical company spun off from ByteDance, completed financing of about $290 million at a valuation of about $1.5 billion. It mainly develops technologies for biomolecular structure prediction, antibody design and drug discovery. Unlike Anthropic, which incorporates drug R&D into its own business, ByteDance chose to spin off the related business and rely on external financing to support long-term development. This also reflects the differences between AI pharmaceuticals and internet businesses in R&D cycles and operating models.

An important reason for tech companies to enter the pharmaceutical industry is that traditional drug R&D has huge room for efficiency improvement.

A Deloitte study on returns from pharmaceutical innovation and R&D released in 2026, which tracked and analyzed the late-stage drug R&D pipelines of 20 large biopharmaceutical companies worldwide, found that in 2025 the average R&D cost for a single drug from discovery to market reached $2.67 billion, higher than $2.23 billion in 2024. The pipeline's projected internal rate of return (IRR) rose from 5.9% in 2024 to 7% in 2025, but the improvement was mainly driven by GLP-1 weight-loss and diabetes drugs. Excluding such drugs, the projected return rate was only 2.9%.

Judging from the recent moves of model companies such as Anthropic and OpenAI, biomedicine and even the broader AI for Science are becoming a key exploration direction after programming.

As for why AI giants are rushing into manufacturing, a relevant person from XtalPi told a Yicai reporter that although large-model companies such as Anthropic have strong AI capabilities, they still face two major shortcomings: first, they lack high-quality proprietary experimental data that is difficult to obtain through public channels; second, the drug R&D process is complex, and computing and predicting molecules is no longer the industry bottleneck, but subsequent laboratory work is still needed for synthesis, testing and validation, which is exactly what pure AI companies lack. It is difficult for them to complete the R&D loop relying only on models, and building a complete experimental system themselves is also unrealistic. Therefore, cooperating with professional institutions that have data, experimental facilities and R&D experience has become an important path for them to enter this field.

From Anthropic to OpenAI, Google and ByteDance, tech companies are entering different stages of drug R&D. But in the end, this competition may not be about who can generate more molecules, but about who can connect models, experimental data, R&D processes and clinical validation to truly improve the efficiency and success rate of new drug R&D.