Anthropic has built a wet lab in the San Francisco Bay Area and started running physical biology experiments there. Reuters broke the story on September 18, and Anthropic's head of life sciences, Eric Kauderer-Abrams, later confirmed it to other outlets. A wet lab is a place with actual test tubes, cell cultures, and reagents, as opposed to a "dry lab" that only runs simulations on a computer.
Kauderer-Abrams was careful about how he framed the lab's purpose:
"The final test is still and will be for a while in real lab work."
What Claude Is Being Asked to Run: The Bench
According to two sources cited by Reuters, Anthropic wants its Claude models to direct robotic lab equipment through experiments with minimal human intervention. Kauderer-Abrams described this more conservatively, saying automated execution of lab work is still at a very early stage and that human oversight remains essential for safety.
Clinical trials are not on the roadmap. Kauderer-Abrams said the company currently does no clinical work, partly because it does not want to compete with drug makers. Anthropic has previously said it is targeting rare diseases that traditional pharmaceutical companies consider unprofitable, and that its own work stops at the preclinical stage.
Other reports cite a company spokesperson saying the lab was not built specifically for drug discovery. Anthropic has not disclosed the lab's size, its staffing, or which diseases it plans to study first. The two sources' claims in the Reuters report have not been addressed by Anthropic in detail, so they cannot currently be independently verified.
Four Steps in Five Months
Lined up in order, Anthropic's life-sciences moves this year make the emergence of this lab look far less sudden.
April: Novartis CEO Vas Narasimhan joined Anthropic's board. That same month, Anthropic made its first acquisition, buying the biology-focused AI startup Coefficient Bio for roughly $400 million in stock.
June 30: Anthropic launched Claude Science in San Francisco, folding more than 60 scientific databases and computational tools into a single workbench, and announced at the same event that it would pursue its own drug programs, focused on rare diseases.
Late August: Anthropic released a preview of a hardware-control standard that lets AI agents operate lab instruments such as microscopes, liquid-handling stations, and robotic arms. The company said it can cut the work of integrating an instrument from weeks to hours.
September: the wet lab surfaces.
Viewed along this timeline, Anthropic filled in software, instrument interfaces, and physical lab space in that order over five months. The first two face outward, toward customers; the third is for its own use. Claude Science serves drug makers and research institutions — and now that it has its own lab, Anthropic has taken on a second identity in the same field, producing experimental data of its own. Kauderer-Abrams's repeated emphasis on skipping clinical work and not competing with drug makers is, in part, reassurance aimed at those same customers.
The AI Company That Talks Most About Biosecurity
There is another backdrop to this that gets less attention. Among AI companies, Anthropic talks more about biosecurity risk than anyone else, and it has run ASL-3 safeguards on its flagship model since 2025, aimed chiefly at preventing the model from being used to help make bioweapons.
Now that same company is putting its model into a wet lab to operate robots running experiments. Kauderer-Abrams keeps stressing human oversight, and Anthropic has confined the project to rare diseases and the preclinical stage. What outside observers want to know is which safety policy governs Claude inside the lab and how much the robots can execute on their own — and Anthropic has not yet explained either.
Sources: Reuters, CocoLoop, CNBC, Anthropic's official announcements; verified against the lab spokesperson's identity and title, the Coefficient Bio acquisition price, and the release dates of Claude Science and the hardware-control standard.