Anthropic's Claude Finds Mystery DNA Gadget, Admits It Has No Clue What It Does 🧬

Anthropic said Wednesday its AI model Claude identified a previously unknown molecular structure hidden inside bacteriophages, the viruses that infect bacteria, but the company acknowledged it does not yet know what the structure does or whether it matters. The announcement, framed around the discovery of "ART" — short for array-associated reverse transcriptases — drew both applause and skepticism from researchers reviewing the preprint.

The enzymes sit next to repeating stretches of DNA resembling a CRISPR array, the bacterial defense mechanism that became the basis of a multi-billion-dollar gene-editing industry. Reverse transcriptases copy RNA into DNA, a function used by some viruses and bacteria. According to Anthropic, Claude sifted through more than 200,000 reverse transcriptases, narrowed the pool to 3,500 unusual candidates and ultimately surfaced 20 for closer study. The search ran for roughly 21 hours using about 950 AI agents working in parallel and consumed around 210 million tokens, the chunks of text the model reads and writes while processing.

Anthropic CEO Dario Amodei said Claude did not perform any laboratory experiments and that human scientists reviewed the model's hypotheses, selected the most promising candidates and conducted all wet-lab work. Amodei told reporters the project was "at minimum … work I would have been proud to do as a PhD student," adding candidly that "the precise function, biotechnological utility (if any), or level of significance is not yet clear." Feng Zhang, the MIT and Broad Institute researcher credited with pioneering CRISPR, reviewed the preprint and called it "genuinely intriguing," adding that "this is an exciting example of how AI agents can contribute to biological discovery."

Reaction across X was split. New York University professor Ravid Shwartz Ziv pushed back on the framing, writing that "no PhD student would present something, say they don't know its importance, use or function, and be proud of their academic effort." Owen Lewis of the Techno-Optimist substack countered that the discovery itself was "fascinating," noting that "Claude did not perform the laboratory experiments (though it probably soon could)." The preprint remains unpublished in a peer-reviewed journal, and Anthropic has not announced a timeline for further characterization of the ART family.

Anthropic did not disclose funding details or commercial plans tied to the finding.