AI · Sep 19, 2026
Vals, backed by Andreessen Horowitz, is looking to become the gold standard for AI benchmarkingAnthropic is operating a lab that conducts biology experiments
Sep 18, 2026, 4:13 PM · TechCrunch

Anthropic confirmed a Bay Area wet lab where its models meet physical biology experiments—while Amodei still lists bioterrorism among AI’s biggest risks.
Why it matters
Julie Bort’s TechCrunch report confirms Anthropic runs a wet biology lab in the Bay Area so its models can test theories in physical experiments. Life sciences head Eric Kauderer-Abrams told Reuters the final test in biology remains real lab work, and that Anthropic is doing that today, partly in-house and partly with external partners.
The lab’s stated focus is fundamental biology, not drug discovery—Anthropic does not want to look like it is competing with pharma customers, including a newly announced Novo Nordisk collaboration. Anthropic bought Coefficient Bio, a stealth AI biotech, in April, and this week launched a Life Sciences Verification Program for vetted researchers plus reports on protein design and biomolecular modeling.
The timing collides with Jacob Coxon’s resignation warnings, Anthropic’s alignment lead putting greater than 10% odds on AI extinguishing humanity within a decade, and Amodei’s weekend call to slow down—while he has repeatedly named bioterrorism as a top AI risk.
From the desk
We’re for AI that accelerates real biology when the experiments are controlled, disclosed, and separated from weaponizable paths—and we hear why the juxtaposition makes people flinch.
Curing disease needs wet labs. Amodei’s five-to-ten-year cure claim is empty without physical validation. Confirming the lab is honesty; calling the focus “fundamental biology” to protect pharma partnerships is also strategy.
The hard part is governance: a lab that can validate helpful hypotheses can also validate dangerous ones. Amodei’s bioterrorism warnings and the wet-lab build are not automatically contradictory—capability without verification is worse—but critics like Chamath Palihapitiya’s tongue-in-cheek post capture the brand whiplash of “regulate me / we’re all going to die” next to SF glassware.
Useful AI in life sciences is a public good if access, dual-use screening, and partner oversight are real. I’m watching what the Verification Program actually gates, what the wet lab publishes, and whether independent biosecurity reviewers get the same embedded access Accenture is getting on models.
Context
TechCrunch published on September 18, 2026. Anthropic declined specifics on current wet-lab projects beyond the fundamental-biology framing.
Who feels it
- Biotech partners
- Clarify whether Anthropic stays infrastructure-and-research or edges into competitive discovery.
- Biosecurity community
- Demand dual-use review commensurate with Amodei’s own bioterrorism risk framing.
- Policymakers
- Wet-lab + frontier-model combos belong in any embedded-evaluator and incident-reporting regime.
What to watch
- Details of the Life Sciences Verification Program’s eligibility and model access tiers.
- Published wet-lab results versus continued secrecy on project lists.
- Whether independent biosecurity orgs join Anthropic’s embedded-evaluator rollout.
Companies: Anthropic