Anthropic Builds Wet Lab to Put Claude in Charge of Drug-Discovery Robots

Anthropic is moving its drug-discovery ambitions off the screen and onto the bench. The AI company is building an in-house biology laboratory where its Claude models will guide robotic systems through physical experiments, according to a report from The Decoder. The move pushes the company past computational screening and into wet-lab work, where compounds are actually mixed, tested and measured.

The setup described is a closed loop: Claude proposes an experiment, robotic hardware carries it out, instruments read the results, and the model uses that data to choose the next step. That cycle is the part labs have struggled to automate at scale, because most AI-driven discovery still ends at a predicted molecule or a simulation that a human then has to validate by hand. Anthropic's lab would shrink the gap between prediction and physical evidence.

Anthropic has never been a drug company, so the build-out marks a strategic shift. Its revenue comes from selling model access, and its research arm has focused on model behavior, safety and interpretability. Standing up a biology lab means hiring scientists who can run assays, buy or integrate robotic liquid handlers and analytical instruments, and manage the regulatory and safety obligations that come with handling biological material.

The company is not alone in chasing this. Recursion Pharmaceuticals has spent years pairing automated labs with machine learning, and Isomorphic Labs, spun out of Google DeepMind, has pursued AI-first drug design with its own partnerships. What differs here is the pairing of a frontier model developer with its own wet lab, which would let Anthropic tune Claude on experimental feedback that outside customers cannot easily supply.

If it works, the payoff is speed. Iterations that once took weeks of human bench work could run continuously, with the model adjusting conditions overnight and flagging compounds worth a closer look. The risks are equally real: automated experiments can generate plausible-looking results that do not replicate, and a model with control over physical equipment raises safety questions Anthropic has said it takes seriously.

The obvious question is what Anthropic intends to do with any discoveries. Running a lab is expensive and far from its core business, so the effort could serve as a proof point for selling to pharma partners, or as an internal pipeline that never leaves the building. Watch for hiring in biology and automation roles, and for whether the company publishes experimental results rather than only model benchmarks.