Anthropic Says Its Biology Lab Has Found a Potentially Major Enzyme System
Anthropic reports that its Bay Area wet biology lab, using its Claude AI model, has identified a new enzyme system with CRISPR-like properties hidden in bacteriophage DNA. The finding still needs validation from the broader research community.
Anthropic recently confirmed it operates a wet biology lab in the Bay Area, where AI models are used to help run physical experiments. This week, the company announced what it describes as a significant discovery made at that lab.
According to Anthropic, researchers identified a previously unknown enzyme system embedded in the DNA of bacteriophages, viruses that infect bacteria. The company says this system has properties reminiscent of CRISPR, the naturally occurring bacterial immune mechanism that has become a widely used gene-editing tool, and that it can perform operations such as cutting, copying, and pasting DNA.
Claude's role in the discovery
Anthropic CEO Dario Amodei acknowledged on X that the discovery builds on prior work by others, noting that a Stanford team had previously found a system that is in some ways similar. Still, Anthropic emphasizes that the new system was found mostly, though not entirely, by its Claude model. The lab has been operating since this spring, and Claude reportedly spent about 21 hours of concerted effort on the discovery, deploying roughly 950 agents and consuming 210 million tokens.
Notably, all physical lab work is carried out by human scientists rather than the AI itself. Anthropic says the lab only handles the lowest biosafety risk levels and does not work with pathogens capable of infecting humans.
Context and risks
The disclosure comes shortly after several AI company leaders, including Amodei, publicly acknowledged that models have grown capable enough to warrant slowing down and developing formal safety-testing procedures. Amodei has previously named AI-enabled bioterrorism as one of his biggest fears, while also stating his belief that AI could help cure most diseases within five to ten years.
Anthropic is not alone in pursuing AI-driven biology research. Stanford researchers recently published work on combining large language models with CRISPR, and UC San Francisco scientists have used AI to design enzymes from scratch. Amodei has not ruled out a future in which Claude autonomously operates lab equipment under appropriate safeguards, though he said that is not happening today.
