Anthropic says its AI Claude found a new enzyme system in phage DNA

Anthropic has announced that its AI system, Claude, identified a previously unknown enzyme system within the DNA of bacteriophages, a finding the company believes could signal a new gene-editing mechanism. The discovery was made public by chief executive Dario Amodei.

According to the company, the enzyme's gene is accompanied by a long array of repeating DNA, a structure that bears some resemblance to CRISPR, the widely used technology for selectively modifying the DNA of living organisms.

Anthropic said the system's function is not yet understood, but noted that only a small number of known systems share its characteristics, and all of them are capable of cutting, copying and pasting DNA. The system is built on a reverse transcriptase, an enzyme that copies RNA into DNA. While that underlying reverse transcriptase, found in a jumbo phage, had been identified in earlier studies, the company said Claude appears to be the first to notice the system's defining features: an associated array of non-coding DNA sequences and an additional accessory protein whose role is unknown.

Amodei described how the work unfolded. The company's life sciences team proposed a broad area of research, after which Claude reviewed literature and genome data, flagged something of interest, and then suggested experiments to verify the finding.

He argued the result should not be treated as an isolated curiosity, pointing to a recurring pattern in which AI performance in new intellectual fields moves from weak to superhuman within a few years. He said AI for biology appears to be on a similar exponential path, and that humans can collaborate with AI to run experiments and validate key results within weeks, iterating as needed. In time, he added, it may be possible for Claude to safely conduct experiments itself by autonomously controlling laboratory equipment, provided suitable safeguards are in place.

Amodei said the work could deepen biological understanding and help uncover new therapeutic approaches, including by identifying drug targets, enabling more precise measurement and shortening the experimental cycle. He said he believes such systems are only beginning to be found and developed into biotechnology tools.

Anthropic said it hopes the work illustrates the value of AI-driven hypothesis generation for the broader scientific community, and that it wants to collaborate with other scientists to apply the approach to a wide range of problems in genomics and beyond.