Anthropic confirmed it operates its own biology lab, where it tests whether Claude can autonomously control robots to run experiments

Anthropic confirmed it operates its own biology lab, where it tests whether Claude can autonomously control robots to run experiments

On September 18, 2026, Eric Kauderer-Abrams, the head of Anthropic's life sciences division, confirmed in an interview with Reuters that the company owns and operates its own wet lab, a facility equipped with test tubes for real experiments, in the San Francisco Bay Area. There, Anthropic is testing whether Claude can control robots and carry out experiments with minimal human involvement; a human presence remains a mandatory safety requirement. A company spokesperson clarified that the lab focuses on fundamental biology, not drug development.

Until now, Anthropic's role in biology was that of a tool provider for other people's labs. In April, the company acquired the startup Coefficient Bio for $400 million in stock. In June, it launched Claude Science, a working environment for researchers, and announced preclinical programs targeting diseases that are financially unattractive to pharma. In August, Claude designed 1320 miniproteins for 15 targets, and their binding was verified by two external contract laboratories: Anthropic itself never touched a test tube. Around the same time, the company introduced the Model Hardware Standard, a way for AI agents to remotely read instrument outputs and adjust settings on third-party equipment; in one pilot, an agent used it to operate lab robots, speeding up experiments by nearly threefold. Now Anthropic has taken the next step: its own lab instead of someone else's, its own experiment instead of a paid verification.

Reliance on external labs was slowing the transition from Claude's predictions to real-world validation, a bottleneck already exposed by the August miniprotein campaign. "There are things that we can do much faster in our own hands," Kauderer-Abrams said, "with the goal of working at the largest possible scale." The decision also reflects organizational growth: the team from the Coefficient Bio acquisition has expanded to the point where the life sciences division is, in his words, among the most fully staffed in the company. "We believe that for biology, the final check will remain real laboratory work for a long time to come," Kauderer-Abrams says, "and our approach is typical of most biotech companies: we do some of the work ourselves and some with external partners."

Anthropic has deliberately limited its ambitions: it does not conduct clinical trials and does not want to be seen as a competitor to pharma and biotech companies that bring drugs to market. One motivation is the trust of pharma clients and partners such as Ozempic manufacturer Novo Nordisk: according to CNBC sources, the company is concerned that clients might suspect it of using data about their competitors' development programs, even though Anthropic keeps such data isolated. For Anthropic CEO Dario Amodei, this commitment is personal: by his own account, illness killed his father just a few years before a treatment for it was found.

The confirmation came at a time when Anthropic itself was publicly weighing the risks: the company reported blocked attempts to use its models for bioweapon development. Shortly before, Amodei had stated, "I believe AI is capable of curing most serious diseases in the next 5 to 10 years," and soon after published an essay calling on the industry to slow down.

The coincidence was noted within the industry as well: investor Chamath Palihapitiya wrote on X, "The team behind such hits as 'We're all going to die' and 'Regulate me immediately' is building a bio lab in San Francisco. I wouldn't recommend it."