Anthropic's newly established wet biology lab has identified a previously unknown enzyme system hidden in bacteriophage DNA that shares key characteristics with CRISPR, the company announced this week. The discovery, which Anthropic CEO Dario Amodei says was made "mostly, though not entirely, by Claude," represents what the AI firm believes is a significant breakthrough in molecular biology. The find centers on an enzyme system capable of manipulating DNA in ways that echo how bacteria use CRISPR to defend against viral attacks.
Claude completed the discovery work in just 21 hours of focused effort, deploying roughly 950 agents that processed 210 million tokens while searching through data. The lab itself has only been running since this spring, meaning the breakthrough came within a few months of the facility's launch. The enzyme system was found tucked away in the genetic code of bacteriophages, which are viruses that invade and multiply inside bacterial cells. All physical experiments at the Bay Area facility were conducted by human scientists, with the lab handling only lower biosafety risk levels BSL-1 and BSL-2 and avoiding any pathogens that can infect people.
According to Anthropic, the newly identified system possesses "properties reminiscent of CRISPR" and can "perform operations like cutting, copying, and pasting DNA." Amodei acknowledged that a Stanford team had previously found a system "that is in some ways similar to the one Claude found," noting the discovery built upon earlier work. The CEO hasn't dismissed the possibility of full automation down the line, stating that "eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren't doing that today."
The announcement arrives amid heightened scrutiny over AI safety in biological research. Amodei and other AI executives recently called for the industry to slow development and establish safety-testing protocols, given how capable and potentially dangerous models have become. Amodei himself has publicly identified bioterrorism as one of his greatest AI-related fears, even as he's predicted AI will "cure most diseases in 5-10 years." The tension reflects a calculated gamble that the medical benefits outweigh the security risks. AI-driven biology research has already expanded beyond dedicated AI companies—Stanford researchers just published work on using large language models with CRISPR, UC San Francisco scientists deployed AI to design enzymes from nothing, and Google launched its AlphaFold biology tool back in 2020.
The broader research community will ultimately determine how novel and important this enzyme system proves to be. For now, Anthropic is emphasizing not just what was discovered but how quickly Claude reached the finding, suggesting AI models can dramatically accelerate the pace of biological research even when humans conduct the hands-on work. The company appears willing to navigate the biosecurity concerns in pursuit of transformative medical applications, while keeping autonomous experimentation off the table for the moment. Whether speed-running biological discoveries creates more opportunities than vulnerabilities will depend on how the industry balances innovation against the safeguards Amodei says are necessary. The bet assumes that human oversight at the lab bench provides sufficient protection even as AI handles the intellectual heavy lifting.