Anthropic built a secret biology lab — and relaxed its own bioweapon safeguards
Anthropic has quietly set up a physical "wet lab" — a facility for hands-on biological experiments — in the San Francisco Bay Area, Reuters confirmed. The company best known for its Claude AI models is now handling real biological materials, not just running simulations. The move marks a significant escalation in Anthropic's push into drug discovery, and raises pointed questions about who is watching the work.
The lab
Eric Kauderer-Abrams, Anthropic's head of life sciences, confirmed the facility exists but gave little else away. Asked what the lab is actually for, he said flatly: "Not drug discovery." What it is researching remains undisclosed. Anthropic declined to reveal the lab's size, headcount, or — most critically — its biosafety level, the classification that determines which pathogens researchers are permitted to handle. Kauderer-Abrams acknowledged that full lab automation is still "very early innings," and that external partnerships remain the company's primary research channel. The in-house lab, he said, supplements rather than replaces those outside collaborations.
The business
The lab is one piece of a broader biotech build-out. In April 2026, Anthropic acquired biotech startup Coefficient Bio for roughly $400 million in stock. In June, it launched Claude Science — a specialized research tool now used by Genentech, Bristol Myers Squibb, and Novo Nordisk, per CNBC. Novartis CEO Vas Narasimhan has joined Anthropic's long-term benefit trust board. Founder Dario Amodei has spoken publicly about a personal stake in the work: his father died from a disease before a cure was found, and he has said Anthropic is targeting "neglected" conditions that traditional pharma avoids.
The safety paradox
Here is the tension. Anthropic has spent years publishing warnings about AI-enabled bioweapon risks — CEO Amodei has described a "substantial risk" of AI-assisted bioterrorism within two to three years. Yet the company relaxed biological safeguards in its Claude Fable 5 model. In the nine months between December 2025 and August 2026, Anthropic documented five biology-related misuse attempts involving dual-use pathogens including avian influenza and orthopoxviruses, according to ScienceTimes. Asked directly how the company balances advancing dangerous capabilities against the risks they create, Kauderer-Abrams offered a candid three-word answer: "We constantly balance."
No FDA or HHS pre-approval of Claude Science for drug discovery has been reported. There is no disclosed timeline for human clinical trials from Anthropic's internal program — putting it behind Google's Isomorphic Labs, which has pursued clinical goals for years, albeit with its own delays.