Anthropic Built Its Reputation On AI Software. Now It Is Running Biological Experiments In Its Own Lab.

Anthropic has established its own biology laboratory in the San Francisco Bay Area, taking the AI company beyond computer-based research and into physical experiments as it expands its work in life sciences.

The facility is a wet lab, where researchers can conduct experiments involving biological materials rather than limiting work to computer simulations. Two people familiar with the operation disclosed the lab to Reuters. Anthropic's head of life sciences, Eric Kauderer-Abrams, confirmed that the company is conducting physical laboratory work.

"We believe that to do biology, the final test is still and will be for a while in real lab work," Kauderer-Abrams said. Anthropic is conducting some of that work at its own facilities while relying on external partners for other experiments, he added.

An Anthropic spokesperson clarified that the lab itself is not specifically a drug-discovery facility. The company has, however, laid out plans to conduct preclinical work on diseases and research areas that traditional pharmaceutical companies may not pursue because of limited commercial incentives.

Anthropic is also exploring how Claude can direct robotic laboratory equipment with limited human intervention, one person familiar with the work told Reuters. The company said human oversight remains essential, while Kauderer-Abrams described laboratory automation as being in its early stages.

The physical lab builds on a broader expansion into scientific research that Anthropic has undertaken this year. In June, the company launched Claude Science, an AI workbench that combines commonly used scientific databases, software packages and computing tools in one environment. The product is designed to help researchers with tasks ranging from literature searches and data analysis to computational research.

Anthropic moved further into physical automation in August with its Model Hardware Standard, a specification designed to allow AI agents to communicate with and operate scientific and manufacturing equipment. The system can connect agents to equipment including microscopes, liquid handlers and robotic arms, with Anthropic initially making it available to selected scientific laboratories and advanced manufacturers.

The company's biology work has also produced experiments involving protein design. Anthropic said in August that Claude models designed protein binders against 15 targets and produced successful binders for 14 of them during tests. Individual designs succeeded at rates ranging from 22% to 35%, depending on the setup, compared with what Anthropic described as typical rates of 10% to 15% in protein-design campaigns.

Anthropic's expansion into life sciences is taking place alongside partnerships with established drugmakers. Novo Nordisk agreed this week to use Claude in drug research and development, adding the Danish pharmaceutical company to a group of drugmakers working with Anthropic's technology. The agreement is aimed at using Claude for scientific research and software development, Reuters noted.

The company has also worked with pharmaceutical groups including Roche's Genentech and Bristol Myers Squibb. In April, Anthropic added Novartis CEO Vas Narasimhan, a physician-scientist who has overseen the development and approval of dozens of medicines, to its board.

Anthropic has expanded its internal life-sciences operation as well. The company acquired biotech AI startup Coefficient Bio, whose team was developing technology for tasks including planning drug research, managing regulatory strategy and identifying drug opportunities. The Information reported that the deal was worth roughly $400 million in stock; Anthropic confirmed the acquisition but did not disclose its value.

The push into physical biology comes as Anthropic is also tightening safeguards around the use of advanced AI for biological work. The company has acknowledged that increasingly capable models can provide information relevant to biological-weapons development, even as it promotes Claude's use for legitimate scientific research.

Anthropic has restricted access to some of its most capable models for professional biology and drug-development queries because of those dual-use risks. When it expanded support for scientists in August, the company said researchers working in biology and chemistry would remain limited to its Opus-class models while it worked with the U.S. government on controlled access to more capable systems.

Those concerns have become part of a wider debate inside Anthropic over the pace of AI development. CEO Dario Amodei recently called for leading developers to coordinate on slowing the advancement of their most powerful systems, while researchers at Anthropic and other major AI companies have issued warnings about the risks posed by increasingly autonomous models.

Anthropic is nevertheless continuing to expand Claude's role in scientific work. Kauderer-Abrams told Reuters that life sciences is already one of the company's largest areas of investment by headcount and resources, with the company particularly interested in diseases considered "undruggable" or neglected by existing drug developers.

The company does not plan to take experimental medicines through human clinical trials. Kauderer-Abrams said Anthropic has drawn that boundary partly because it does not want to compete directly with pharmaceutical and biotechnology companies whose business is bringing medicines to market.