Anthropic says its biology lab already made a big discovery

Anthropic confirmed last week that it operates a wet biology lab in the Bay Area, where it uses its AI models to run physical experiments. This week, the company said the lab has already made what it believes is a major discovery: a previously unknown enzyme system hidden in the DNA of bacteriophages — viruses that infect and replicate within bacteria — with properties Anthropic likens to CRISPR, the natural bacterial immune system that has become a widely used gene-editing tool. Anthropic’s researchers say the new system can carry out DNA operations similar to cutting, copying, and pasting, much like CRISPR does.

It will be up to the broader research community to validate how significant, or how new, the discovery actually is. Anthropic CEO Dario Amodei has acknowledged that the find builds on the work of others, writing on X that a Stanford team had previously discovered a system he described as in some ways similar to the one Claude found. Even so, Amodei and Anthropic emphasize that this discovery was made “mostly, though not entirely, by Claude.” The lab was established only this spring, and the team says the search itself took Claude 21 hours of concerted effort, using roughly 950 agents that burned through 210 million tokens.

The disclosure follows AI CEOs, including Amodei, publicly acknowledging that their models have grown capable — and potentially dangerous — enough that the industry needs to slow down and build safety-testing procedures, after some of Amodei’s own employees said earlier this month that AI carries a risk of killing everyone. Amodei has said one of his biggest fears is AI being used for bioterrorism, though he also believes AI could help cure most diseases within 5-10 years.

Notably, the lab hasn’t let Claude run its own physical experiments: all lab work is performed by human scientists, and the lab handles only lower-risk biosafety levels (BSL-1 and BSL-2), with no pathogens capable of infecting humans. AI-powered biology research isn’t unique to Anthropic either — Stanford researchers have published on LLMs and CRISPR, UC San Francisco researchers have used AI to design enzymes from scratch, and Google launched its AlphaFold biology tool back in 2020. Still, Amodei hasn’t ruled out a future where Claude autonomously controls lab equipment to run experiments itself, with appropriate safeguards, though he says that isn’t happening today.

Why it matters

The episode shows AI labs racing to claim scientific breakthroughs even as their own leaders warn the same models could pose catastrophic risks — a tension Anthropic is managing by keeping its AI out of the physical lab work for now.