Morning Edition · Thursday, September 24, 2026Published at 2:18 AM EDT · New York
The run used 210 million tokens to screen more than 200,000 reverse transcriptases down to 20 candidate systems, and the biological function of the top candidate is still unknown.

Anthropic reported that an agent system built on its Claude model identified a previously undescribed enzyme family, which it calls array-associated reverse transcriptases (ART). These enzymes sit next to long arrays of evenly spaced DNA repeats that resemble the pattern known as clustered regularly interspaced short palindromic repeats (CRISPR). The company published the result on Tuesday.
The scale of the operation matters as much as the biological finding. Roughly 950 agents ran for 21 hours and used 210 million tokens, according to a summary circulated by the Russian-language channel AI ML Big Data that matches Anthropic's own account. The agents collected more than 200,000 reverse transcriptases (the enzymes that copy RNA into DNA), flagged about 3,500 candidate systems, and produced detailed reports on 20 of them.
What has actually been verified is limited. The ART system combines a reverse transcriptase, a partner gene, and a repeat array, a combination that so far appears only in a small number of known systems, all of which are programmable and can cut, copy, or insert DNA. Whether the ART system performs any of those functions inside a living cell has not been demonstrated. No peer-reviewed paper accompanies the announcement. The strongest outside endorsement is a supportive comment from Feng Zhang, a researcher at the Massachusetts Institute of Technology and the Broad Institute, quoted in coverage of the release, who called it an example of AI agents contributing to biological discovery.
Anthropic has a direct commercial interest in the claim holding up, and the company has framed the work as a computational search of sequence data rather than laboratory validation. The most accurate description is that Claude generated hypotheses at a scale and speed no human team could match by searching public sequence databases, and that the biology still needs to be confirmed through laboratory experiments.
Part of a tracked trend
AI Moves Into Autonomous Scientific Discovery and Clinical Care
Over the next 3-9 months, AI systems move beyond text tasks into running real scientific experiments and managing clinical care, backed by peer-reviewed and benchmarked evidence of chemist- and physician-level performance.
Start a discussion in Townsquare.
More from this edition
Anthropic, which needs a scientific-discovery proof point to justify enterprise agent pricing, and the biotechnology investors who would hold the intellectual property if the array-associated reverse transcriptase system turns out to be programmable.
The article understates two disclosures: the reverse transcriptase itself was already known and Claude's contribution was noticing the adjacent repeat array and partner protein, and Anthropic reran the same campaign ten more times and every rerun missed the array, so the result is a preprint awaiting peer review rather than a reproducible method, while "CRISPR-like" describes structural resemblance only.
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
Agent computing time is becoming a resource that research groups can buy for discovery work, and 210 million tokens is now a cost a biotechnology company can budget against the cost of a graduate student's year of labor. If laboratory work confirms that the ART system is programmable, the winners are the owners of large sequence databases and labs that can pair agent-driven search with fast experimental validation. The losers are contract research organizations that sell human-driven screening. If the biology does not hold up, the search method itself remains valuable and will likely be applied to other poorly studied protein families.
What to watch
Observations to monitor, not financial advice.
Synthesized from: Anthropic · Polylog editors
Comments
0No comments yet.