On September 23 2026, Anthropic’s life-sciences team reported that its AI model Claude uncovered a previously unknown enzyme system resembling CRISPR, dubbed array-associated reverse transcriptases (ART), in bacteriophages, after mining DNA data and confirming the find with bench experiments.
Short answer: On September 23 2026, Anthropic’s life-sciences team reported that its AI model Claude uncovered a previously unknown enzyme system resembling CRISPR, dubbed array-associated reverse transcriptases (ART), in bacteriophages, after mining DNA data and confirming the find with bench experiments.
On September 23, 2026, Anthropic’s life sciences squad revealed that its AI model Claude had dug up a previously unknown enzyme system that bears a striking resemblance to CRISPR. The find sprang from a brand-new lab and research group put together earlier that spring, whose charter is to put general-purpose AI to work on fundamental biology-scouring DNA datasets, spinning up hypotheses, and then putting those ideas to the test in the wet lab. By marrying AI-driven data mining with classic bench work, the team’s first results show how this partnership can surface fresh biological mechanisms.
Claude was set loose on a massive trove of DNA sequences, hunting for interesting reverse transcriptases-the enzymes that copy RNA back into DNA. Humans only had to write the initial prompt and run the follow-up lab tests; the AI did the heavy lifting, sifting through distinct reverse transcriptase families and using its own judgement to flag promising candidates. In roughly twenty-one hours, about nine hundred fifty Claude agents chewed through roughly two hundred ten million tokens of sequence data. One agent spotted a repeating DNA motif sitting next to the gene for an unusual reverse transcriptase, which triggered a deeper dive.
After that first glimpse, Claude went about its business much like a human scientist would when faced with a potential breakthrough. It counted the repeats, measured their spacing, stacked the arrangement up against known reverse transcriptase systems, and combed the literature for any prior mention of the pattern. Convinced it had stumbled onto something unseen, the AI shipped a report off for human review. Bench experiments confirmed that the motif marked a previously uncharacterized enzyme system living inside bacteriophages-those viruses that prey on bacteria. The researchers christened the discovery array-associated reverse transcriptases, or ART.
ART is built around a reverse transcriptase first seen in a jumbo phage, but Claude’s analysis uncovered two extra pieces that had never been linked before: an adjacent array of non-coding DNA repeats and an accessory protein whose job is still a mystery. Although we haven’t nailed down ART’s exact biological role yet, the combo of these traits shows up in only a handful of other systems, all of which are known to be programmable and able to perform DNA operations such as cutting, copying, or pasting. CRISPR, the poster child of this crowd, has already reshaped genetics and medicine, and a few similar mechanisms are currently being tweaked as possible tools for future applications.
The episode really puts Claude’s scale on display. During its investigation the AI looked at more than two hundred thousand reverse transcriptase sequences, pulled out thirty-five hundred novel candidate systems, and winnowed those down to the twenty most compelling candidates for a deep-dive report. A human expert would need weeks or months to pull off a survey of that size, yet Claude knocked it out in a fraction of the time. When it zeroed in on the unusual reverse transcriptase family that eventually led to ART, the AI inspected the raw DNA flanking the gene, noticed a tandem repeat array that reminded it of CRISPR-like repeats, counted and spaced those repeats, compared the layout to known systems, and checked existing publications before deciding it had found something new.
Anthropic’s Bay Area facility runs at biosafety levels one and two and doesn’t work with pathogens that can infect people. All the hands-on lab work is done by human scientists, although the team has played around with using AI to speed up certain lab chores through efforts like the Model Hardware Standard-which they found less suited to the improvisational vibe of molecular biology. For data crunching and hypothesis generation, the researchers lean on Claude Science and Claude Code, the same tools any outside scientist can grab, sometimes juiced up with a custom harness that keeps many parallel Claude sessions humming.
The broader scientific community has greeted the news with enthusiasm. Feng Zhang, a heavyweight in CRISPR development and a professor at MIT and the Broad Institute, said the work exemplifies how AI agents can pitch in on biological discovery. He called the identification of RNA-repeat arrays tied to reverse transcriptases genuinely intriguing and worthy of further study, and he voiced hope that the effort will nudge more researchers to treat AI as a partner in their investigations.
Anthropic has dropped a pre-print that lays out the ART system and the methods used to uncover it, inviting others to build on these early clues. While the exact function of ART remains to be elucidated, its discovery undersc
The system is called array-associated reverse transcriptases, or ART. It consists of a reverse transcriptase first seen in a jumbo phage, an adjacent array of non-coding DNA repeats, and an accessory protein whose function remains unknown.
Claude scanned a massive DNA dataset for reverse transcriptases, spotted a repeating DNA motif next to an unusual reverse transcriptase gene, counted and spaced the repeats, compared the arrangement to known systems, checked the literature, and then passed the finding to humans for bench validation.
ART’s exact biological role has not yet been determined, but its combination of a reverse transcriptase, non-coding DNA repeat array, and accessory protein resembles other programmable systems that can cut, copy, or paste DNA, much like CRISPR.
Feng Zhang said the work exemplifies how AI agents can contribute to biological discovery, called the identification of RNA-repeat arrays linked to reverse transcriptases genuinely intriguing and worthy of further study, and hoped it would encourage more researchers to treat AI as a partner.
Source: Anthropic
On September 23, 2026, OpenAI launched MentalHealthBench, an expert-informed benchmark that evaluates how helpful and safe AI responses are in realistic mental-health conversations involving anxiety, depression, stress, and related topics.
California Governor Gavin Newsom signed seven bills on September 21 2026 that require data centers to disclose their electricity and water consumption. The legislation directs the Public Utilities Commission to set separate power rates, mandates monthly energy reporting starting in 2027, obliges operators to report water use when seeking permits or licenses, makes them fund needed water-infrastructure upgrades, and removes a prior exemption so each new project faces standard environmen
California Governor Gavin Newsom signed a package of bills on September 23, 2026 that requires data-center operators to disclose their electricity and water use, mandating monthly energy reporting and water-use reporting tied to permitting or licensing, while setting separate power rates and removing CEQA exemptions.
NO COMMENTS YET
Comments are open. Have a thought or a question? Share it below.