TL;DR
Anthropic has set up a laboratory in the San Francisco Bay Area for physical biology experiments, Reuters reports, and its head of life sciences has confirmed it on the record. The company wants Claude to direct lab robots with limited human involvement, though it says it will not run clinical trials. It marks a move from supplying AI to drugmakers towards doing some of the science itself.
From simulation to the bench
Two sources close to the project told Reuters the lab takes Anthropic’s biology work beyond computer-only evaluation. Eric Kauderer-Abrams, who leads life sciences, said real lab work remains “the final test” and described a typical biotech split, with some experiments run in-house and others through partners. A spokesperson later said the lab is not dedicated to drug discovery, without explaining further.
One source said the aim is for Claude to operate robotic lab units with minimal human intervention. The company’s position is that human oversight stays essential for safety. Kauderer-Abrams called it “the very early innings” of automating lab execution.
How the pieces fit
The lab sits on top of a run of moves this year. At a June event, Anthropic said it would start its own preclinical drug programmes in areas pharmaceutical companies find commercially unattractive. It has launched Claude Science, brought Novartis chief executive Vas Narasimhan onto its board, and confirmed buying Coefficient Bio, reportedly for about $400m in stock. In August it released the Model Hardware Standard for AI control of lab equipment. Kauderer-Abrams said life sciences now ranks among Anthropic’s biggest investments in both staff and resources.
The commercial tension is obvious. Anthropic sells tools to Genentech, Bristol Myers Squibb and Novo Nordisk, which added Claude to its drug discovery stack last week. Two people told Reuters those customers may worry Anthropic could learn from their programmes, even with their data walled off. Staying out of clinical trials is partly a response: “We’re not competing with pharma and biotech companies that make their business in bringing drugs to market,” Kauderer-Abrams said.
Looking forward
The closest comparison is Alphabet’s Isomorphic Labs, based in London, which has pushed back its own target for reaching the clinic until late 2026. Which diseases Anthropic is targeting, and how far that work has got, is still unclear. For UK life sciences firms using frontier models, the practical question is contractual: what their AI supplier can and cannot learn from their data now that the supplier runs experiments of its own. The timing is also awkward, arriving as Anthropic prepares for a public listing and its researchers warn publicly about catastrophic AI risk.