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

Claude is now pointing robot arms at real glassware. Reuters reported on Friday that Anthropic quietly set up a wet lab in the San Francisco Bay Area, and the company confirmed it.
At a glance
- Anthropic has moved part of its science work off the screen and into glassware, with a Bay Area wet lab running physical biology experiments instead of computer-based evaluations.
- Claude directs the robotic units doing the bench work, using the Model Hardware Standard that Anthropic launched in August as the technical framework for a model to operate lab equipment.
- Anthropic says human oversight stays essential to "limit the risk", and no timeline has been given for the neglected-disease programs the new lab is supposed to feed.
If you have not been following this thread: Anthropic's science push has so far lived entirely inside the computer. Models read, reason, score well on evaluations, and nothing they conclude touches a cell. Claude Science is the program being built around that work, and the wet lab is the end of it where an answer gets checked against actual biology rather than against a benchmark.
The target is preclinical work established firms find "financially unattractive"
Eric Kauderer-Abrams, who heads life sciences at Anthropic, said the focus is preclinical research in areas traditional firms find "financially unattractive". That framing also sets the boundary: no direct competition with established pharmaceutical companies.
The lab feeds two things. One is Claude Science. The other is preclinical programs for neglected diseases, the class of illnesses that rarely support a commercial pipeline. Preclinical is everything before a compound goes near a person: cells, assays, animal work, the long grind of showing an idea survives contact with biology.
The Model Hardware Standard, launched in August, is what puts Claude on the equipment
Claude directs the robotic units that run the experiments. The connective tissue is the Model Hardware Standard, which Anthropic launched in August as a technical framework for a model to operate laboratory equipment.
Lab gear is a zoo. Each instrument has its own control software and its own way of reporting what it did. A standard is the printer driver of that world: one agreed way to send an instruction and read back a result, so whatever issues the commands does not need custom plumbing for every machine on the bench. That is the difference between a model that can write a protocol and a model that can run one.
Anthropic maintains that human oversight remains essential, and puts the purpose of it plainly: to "limit the risk" in life-science work.
A reported $2 trillion IPO sits behind the expansion
Anthropic is reportedly preparing for a potential $2 trillion IPO to fund the expansion, and the physical-biology move is part of that build-out rather than a side project.
The comparison the reporting reaches for is Isomorphic Labs, which has pushed its clinic entry to the end of 2026. That slip is cited as a measure of the technical hurdles in AI-driven lab automation. Entering the clinic means a candidate reaching human trials, one full stage past where Anthropic's benches sit. Nothing in the report puts a date on when work from the Bay Area lab would reach that point.
The report leaves a lot unspecified: the size of the lab, how many programs run inside it, which neglected diseases are in scope, and what any of it costs. In our view the "financially unattractive" line is a convenient piece of positioning, because it stakes out preclinical ground where the results will not be graded against an incumbent's pipeline.
Isomorphic's end-of-2026 clinic date
The only dated checkpoint in this story belongs to someone else: Isomorphic Labs has pushed its clinic entry to the end of 2026. Anthropic has no equivalent date on the board. No first preclinical program has been named, no disease, no milestone for when robot-run experiments are meant to produce a candidate. The thing worth watching is whether the Model Hardware Standard turns up on benches outside Anthropic's own.
Related stories
- WHO Africa and CEPI turn to Claude in an Ebola outbreak
- Claude found a CRISPR-like puzzle hidden in phage DNA
- Claude wrote GPU code and made biology models 4x faster
- Mythos opens to verified biologists for the first time
- Ozempic maker adds Claude next to OpenAI's Rosalind
- Anthropic opens an AI for Science track for rare disease research
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.
