Anthropic’s new wet lab puts Claude in charge of robots

Anthropic has opened a wet lab in the San Francisco Bay Area, where Claude directs robotic units running physical biology experiments. Reuters reported the lab on September 18, and the company confirmed it. The lab moves Anthropic’s biology work beyond computer-based evaluations to experiments with physical samples and reagents.

Anthropic had announced the drug program before the lab. At a June 30 event in San Francisco, head of life sciences Eric Kauderer-Abrams said Anthropic would run an internal drug discovery program on “neglected” diseases. He said building the right tools takes tight feedback loops and that “we need to live it along with all of you”, CNBC reported. Claude Science launched the same day. It is an agentic research workbench pitched as doing for lab science what Claude Code did for software engineering, released in beta to paid Claude subscribers with a reported 60-plus functions across genomics, structural biology, proteomics and cheminformatics.

Running experiments needs a layer that software connectors never had to provide. When Anthropic launched Claude for Life Sciences in October 2025, the model reached lab work through connectors to other platforms and Agent Skills, folders of instructions and scripts that keep Claude on a specific protocol. The first scientific skill was single-cell-rna-qc. In the same announcement Anthropic reported that Claude Sonnet 4.5 scored 0.83 on Protocol QA, a benchmark of lab-protocol understanding, against a human baseline of 0.79. The new lab extends that to instruments: the Model Hardware Standard, launched in August, gives the model a technical framework for operating lab equipment.

Anthropic has also spent money on the science side. In April it bought Coefficient Bio, a stealth New York startup with fewer than 10 employees, for a reported $400 million, and Novartis chief executive Vas Narasimhan joined its board.

Kauderer-Abrams describes the target as preclinical research in areas that traditional firms find “financially unattractive”, which keeps Anthropic out of direct competition with established drugmakers. Some of those drugmakers are also its customers, which makes the choice of target convenient. Still, Peptide and Protein News notes that Anthropic is now, in a limited way, a competitor to some of the companies it sells to. The demonstration of its in-house programs at launch involved phenylketonuria, a rare inherited metabolic disorder.

What happens to a promising candidate is still open. In June, Kauderer-Abrams did not say what Anthropic would do with one. A spokesperson said that as a public benefit company, Anthropic can choose programs based on patient benefit, including work the commercial market overlooks.

Separately, Anthropic is preparing for a potential $2 trillion IPO to fund its expansion. Rival Isomorphic Labs, meanwhile, has delayed its entry into the clinic to the end of 2026.

On safety, Anthropic says human oversight of the robotic work remains essential and that it is balancing rapid expansion with careful deployment to “limit the risk” in life-science work. The October 2025 announcement set a longer goal: AI models that eventually make discoveries on their own.