Anthropic has confirmed it operates a wet biology lab in the Bay Area where its AI models run physical experiments, according to TechCrunch AI. Not simulations. Actual benchwork with pipettes, cells and reagents. The company told TechCrunch the lab focuses on fundamental biology rather than drug discovery, and it works both on internal research and with external partners.
Threat assessment first, because that’s how this story is landing. The company whose CEO called bioterrorism one of AI’s biggest risks now owns a facility where an AI helps run biology experiments. Opportunity framing second: if AI is ever going to cure disease, somebody has to close the loop between a model’s hypothesis and a physical result. Anthropic just decided that somebody is Anthropic.
What we know
- Location: Bay Area. Anthropic won’t say more than that.
- Mission: fundamental biology, not drug discovery. That distinction is deliberate, and I’ll get to why.
- Operating model: Eric Kauderer-Abrams, Anthropic’s head of life sciences, told Reuters it runs “like most biotech labs,” with in-house work plus partner projects.
- The quote that matters: “We believe that to do biology, the final test is still, and will be for a while, in real lab work. We absolutely are doing that today.”
- The origin: Anthropic bought Coefficient Bio, a stealth AI biotech, back in April. TechCrunch AI notes this news “probably shouldn’t be shocking” given that acquisition.
Why this matters
The AI industry has been making an enormous promise. Dario Amodei said just last week: “I believe that AI could cure most major diseases in the next 5–10 years.” Nice line. But a language model can’t cure anything from inside a data center. It can propose protein structures, rank candidate molecules, and design experiments. Then a human in a lab coat has to check whether reality agrees.
The status quo until now: frontier labs published papers and shipped tools, and pharma partners did the physical validation. Anthropic has published reports on accelerating protein design and improving biomolecular modeling. Useful, but downstream of someone else’s bench.
A wet lab changes the feedback loop. The model proposes, the lab tests, the result feeds back into the model. Faster iteration, tighter control, and no waiting on a partner’s schedule. What stands out here is that Anthropic is treating biology the way it treats coding: give the model an environment where it can actually run things and see what breaks.
The pharma tightrope
Here’s why the “fundamental biology, not drug discovery” framing isn’t a throwaway. Anthropic sells to pharma. It just announced a joint drug discovery deal with Novo Nordisk. It has already taken heat for launching products that looked like they competed with its own customers. A drug discovery lab would make every pharma partner nervous overnight.
So Anthropic is doing three things at once:
- Running the lab, but keeping it away from drug pipelines.
- Launching a Life Sciences Verification Program this week to give vetted bio researchers access to its most powerful models.
- Publishing research that supports drug discovery without doing the discovery itself.
That’s a fine line. Fundamental biology and drug discovery share equipment, techniques and often the same people. I’d expect pharma partners to watch this closely.
The awkward timing
This lands in a strange week for Anthropic. Researcher Jacob Coxon resigned with a warning that “the people building AI earnestly believe that it could kill us all by the end of the decade.” Anthropic’s own alignment lead has put the odds of AI wiping out humanity within ten years above 10%. Amodei published a post last weekend asking the industry to slow down and self-regulate.
Chamath Palihapitiya summed up the mood on X, half joking: “The group behind such hits as: ‘We’re All Going To Die’ and ‘Regulate Me Now’ are building a wet lab in SF. I do not recommend this.”
The critique is easy to make, but it’s worth separating the two things. A wet lab doing fundamental biology under company control is not the bioterrorism risk Amodei talks about. That risk is about a random bad actor getting expert-level guidance from a model. Still, optics matter, and Anthropic handed its critics a very good punchline.
What to watch
- Whether “fundamental biology” stays fundamental, or drifts toward therapeutics as results come in.
- How OpenAI and Google DeepMind respond. DeepMind already spun out Isomorphic Labs for drug discovery. Anthropic’s move suggests physical labs are becoming table stakes for frontier AI companies with biology ambitions.
- Who gets into the Life Sciences Verification Program, and what “vetted” actually means in practice.
- Biosafety governance. If Anthropic wants credibility here, it needs to publish how the lab is secured, not just what it studies.
My take: this is the right move for the science and a headache for the messaging team. Models that can test their own hypotheses will get better at biology faster than models that can’t. The question is whether Anthropic can keep its safety story intact while running the kind of facility its own warnings make people nervous about. TechCrunch AI has the full report.