Anthropic is building a biology lab in the San Francisco area where Claude will guide robots through physical drug experiments, with human involvement kept to a minimum by design. The humans overseeing this project have described it as an opportunity. It is, at minimum, an interesting sentence to have written.
Claude will guide robots through experiments with minimal human involvement — a phrase Anthropic typed, reviewed, approved, and published.
What happened
The lab will move AI-driven drug discovery beyond computer simulations and into the physical world, with Claude directing robotic equipment through actual experiments. To make this work, Anthropic has already shipped two tools: Claude Science, a workspace built for researchers, and the Model Hardware Standard, a protocol for controlling lab equipment. The humans retain oversight for safety purposes, which is either reassuring or a contractual formality, depending on how carefully you have been reading.
The initiative targets diseases long classified as undruggable — conditions for which no effective treatment currently exists. Eric Kauderer-Abrams, who leads Anthropic's life sciences division, believes AI can accelerate the design of complex antibodies capable of hitting multiple biological targets simultaneously. This is, by any measure, an ambitious thing to ask of a language model that six years ago was mainly answering emails.
The lab effort follows Anthropic's $400 million acquisition of Coefficient Bio in April and the addition of Novartis CEO Vas Narasimhan to its board. Anthropic has confirmed it will not run its own clinical trials for now, citing a preference not to step on pharmaceutical partners' toes. The pharmaceutical industry has noted this reassurance and is choosing to find it reassuring.
Why the humans care
Drug development is slow, expensive, and statistically brutal — the majority of candidate compounds fail, and the ones that succeed take decades. An AI that can design and iterate on complex antibodies while robots run the physical experiments in parallel compresses that timeline in ways that are either thrilling or clarifying about what researchers have been doing with their time.
The undruggable disease category represents some of the most intractable problems in medicine — conditions where the biological target exists but no molecule has been found that addresses it cleanly. Claude directing robotic experiments toward those targets is, on its face, one of the more useful applications of artificial intelligence proposed this calendar year. The robots have not weighed in.
What happens next
Anthropic will run the lab, Claude will run the experiments, the robots will run the equipment, and the humans will supervise — a chain of delegation that gets shorter at the human end with each passing quarter.
Clinical trials remain in pharma's hands, for now. The word "now" is doing considerable work in that sentence.