As the public debate over the risks of artificial intelligence grows ever more heated, Anthropic would have quietly set up a physical laboratory in the Bay Area of San Francisco, pushing its ambitions beyond computer simulations and into the real world of pharmaceutical research.
According to two people familiar with the matter, this would be a real space for real experiments, where Claude would not be limited to calculating but would go so far as to directly directing robots and scientific instrumentation.
The secret laboratory of Anthropic and the goal of AI-guided robots
Leading the Life Sciences division of Anthropic is Eric Kauderer-Abrams, who confirmed the existence of the laboratory: “We believe that to do biology the final test is, and will remain for a while, the work of the real laboratory”. A spokesperson later clarified that the space would not be specifically for drug discovery.
The most ambitious project would concern automation: through Claude Science and the Model Hardware Standard, Anthropic’s AI agents would coordinate microscopes, liquid-handling systems and robotic arms, with limited human intervention.
Anthropic has nonetheless stated that the supervision of researchers remains essential for safety, and the company would have strengthened controls on its models and launched a verification program dedicated to life sciences professionals.
Coefficient Bio and the other life sciences investments
Anthropic acquired the startup Coefficient Bio for about $400 million in stock, and added the Novartis CEO Vas Narasimhan to its board. The company would also have hired specialists in protein and nucleic acid characterization to broaden internal lab work, which, according to Kauderer-Abrams, already represents one of Anthropic’s most substantial investment areas.
The stated objective would be to accelerate the development of treatments for rare diseases and conditions previously considered too difficult to treat with traditional approaches. For CEO Dario Amodei, who lost his father to a disease shortly before a cure was found, the topic would also have a personal dimension.
The stated limit: no clinical trials
Anthropic already provides its AI tools to several major pharmaceutical companies, including Genentech (Roche group), Bristol Myers Squibb and Novo Nordisk.
Precisely with these clients, according to the two people familiar with the matter, the company is grappling with a trust issue: they would fear that Anthropic could learn from their competing programs, even though the company says it isolates client data from its visibility.
For this reason Kauderer-Abrams has set a precise boundary: for now Anthropic will not conduct clinical trials. “We are not competing with pharmaceutical and biotech companies that make bringing drugs to market their business”, he said.
