Anthropic’s Secret Wet Lab: AI‑Driven Biology Experiments Behind the Scenes

When Anthropic last week issued a stark warning about the possibility of AI‑crafted bioweapons, it seemed to be sounding the alarm on a danger it itself was trying to avoid. Yet, as Reuters uncovered, the company has been running a fully automated biology lab in the Bay Area, where its Claude AI talks to robotic assistants that actually pipette, mix and observe samples.

Two anonymous sources, speaking on condition of anonymity, told reporters that Anthropic wants Claude to coordinate the whole workflow – from experimental design to data capture – with almost no human hands on deck. The aim, according to the firm’s life‑sciences chief Eric Kauderer‑Abrams, is to accelerate drug‑discovery‑type research without stepping into the risky territory of clinical trials, where they would be squarely competing with big‑pharma partners.

“We believe that to do biology, the final test is still and will be for a while in real lab work,” Kauderer‑Abrams said to Reuters. “We absolutely are doing that today.” In practice, that means Claude suggests a protein‑folding experiment, a robot arm draws up reagents, a camera watches the reaction, and the AI logs the outcome – all in a loop that can repeat thousands of times a day.

The move isn’t entirely surprising. Anthropic’s CEO Dario Amodei has long spoken about his personal motivation: his father died of hepatitis C just before a breakthrough treatment arrived. In 2024 he famously claimed that “AI‑enabled biology could compress fifty to a hundred years of progress into five to ten.” The new wet lab is a concrete step toward that vision.

But the timing feels odd. On the same weekend Amodei urged the industry to “pause” further AI development, citing the specter of AI‑driven bioterrorism. A few days later, the company quietly confirmed the existence of its own lab, while refusing to detail what exact research is being run there.

Critics argue that the line between drug discovery and bioweapon research is thin – the same synthesis pathways, the same high‑throughput screening rigs, the same data pipelines. If an AI can design a potent molecule, it could also suggest ways to weaponize it. Kevin Esvelt, a synthetic‑biology professor at MIT, recently warned that a chatbot suggested a novel bioweapon he hadn’t considered possible, prompting him to double‑check the claim with Anthropic’s own Claude Fable.

Anthropic says the lab is still in its “early innings,” a phrase that sounds almost like a baseball metaphor for a project that has barely left the batter’s box. Yet the very existence of a robot‑run wet lab underscores how quickly AI is moving from theory to hands‑on biology. Whether that will usher a new wave of medical breakthroughs or open a Pandora’s box of misuse is a debate that will likely continue for years.

For now, the lab hums quietly in a Bay Area building, Claude issuing instructions in a language only robots understand, while the broader AI community wrestles with how, and if, to put the brakes on such capabilities.