Claude helped scientists find a previously unknown biological system hidden inside a massive DNA database
Anthropic, the artificial intelligence company that makes the Claude chatbot, said this week that its AI system had identified a previously uncharacterized enzyme system in viral DNA, a finding the company frames as evidence that AI can accelerate basic scientific research. Independent scientists say the result is intriguing but early, and its practical significance is not yet established.
The company said roughly 950 instances of its Claude model, working as autonomous software agents, spent about 21 hours searching a database of more than 200,000 enzymes called reverse transcriptases, which copy RNA into DNA. One agent flagged an unusual pattern of repeating DNA sequences sitting next to one such enzyme in the genome of a bacteriophage, a virus that infects bacteria. Anthropic named the resulting three-part system array-associated reverse transcriptases, or ART.
The discovery was made by a life sciences research group and laboratory that Anthropic formed in the spring of 2026 and disclosed publicly for the first time this week, alongside a preprint describing the work. Human researchers at the Bay Area lab set the initial search parameters and then carried out physical laboratory testing to confirm the AI-identified pattern was real; the AI agents were not involved in handling lab materials.
Anthropic said the system’s function remains unknown. Initial lab experiments found that the repeat region of ART is expressed as a set of short RNA molecules, a structural feature the company said resembles CRISPR, the bacterial immune mechanism that was adapted into a Nobel Prize-winning gene-editing tool. Anthropic said this arrangement has previously been observed only in a small number of other systems, all of which can be programmed to cut, copy or paste DNA, but it stopped short of saying ART can do the same.
How the search was conducted
According to Anthropic’s account, researchers gave Claude a broad instruction to search the database for unusual examples of reverse transcriptases, then left the agents to work with limited additional guidance. The agents gathered more than 200,000 candidate enzymes, narrowed the list to about 3,500 systems not matching any known category, and produced detailed reports on the 20 most promising candidates — a process the company said would typically take a human researcher weeks or months.
While reviewing raw DNA sequence data near one unusual enzyme, an agent identified a repeating structural pattern, compared it against known systems, checked prior scientific literature for earlier descriptions of it, and referred the finding to human scientists for review, the company said. The underlying enzyme itself had been noted in earlier research, but Anthropic said its agents were the first to identify the surrounding repeat array and an accompanying protein of unknown function as parts of a single system.
Outside reaction
Feng Zhang, a CRISPR gene-editing researcher at the Massachusetts Institute of Technology and the Broad Institute, reviewed the preprint at Anthropic’s request. He said the identification of repeat arrays linked to reverse transcriptases was “genuinely intriguing and merits further investigation.” Anthropic did not identify any scientist unaffiliated with the company or the preprint’s review who has assessed the findings; the preprint has not undergone independent peer review.
Separately, Anthropic chief executive Dario Amodei said on social media that researchers at Stanford University had, independently and using conventional methods, described a related but distinct reverse-transcriptase system with a similar repeat structure, a claim that has not been independently verified and which Anthropic did not include in its own release.
Background
Anthropic’s life sciences group was formed within the company’s broader research organization, which also includes teams working on drug discovery and on training Claude for biology and chemistry tasks. The company said its laboratory operates at the two lowest biosafety containment levels and does not work with pathogens capable of infecting humans, and that all physical experiments are conducted by human scientists rather than by AI systems. Anthropic has previously experimented with AI-assisted laboratory automation through a separate research effort but said that approach was not suited to the exploratory nature of this project.
The company has framed the disclosure as part of a broader argument, made publicly by Amodei, that AI models are approaching the point of contributing to unsolved problems in fields including mathematics and biology. That framing is a claim by the company promoting its own technology, not an independently verified assessment.
What follows
Anthropic said further laboratory work is underway to determine what ART actually does and whether it can be adapted into a usable tool, and that it is inviting outside researchers to propose follow-up studies. The company has not said when results of that further testing might be available.
What remains unconfirmed: whether ART performs any of the DNA-cutting, copying or pasting functions associated with programmable systems like CRISPR; whether the Stanford team’s system, as described by Amodei, is in fact comparable; and how the findings will hold up once the preprint undergoes formal peer review.
Sources
Factual details on the discovery, its methodology and quoted material are drawn from Anthropic’s own announcement and preprint. Casualty-free corroboration of secondary details (agent count, search duration, token usage) comes from multiple independent outlets citing the same company disclosure.
Wire services and international press
- Al Jazeera, “AI model Claude discovers CRISPR-like enzyme system, Anthropic says.” https://www.aljazeera.com/economy/2026/9/24/ai-model-claude-discovers-crispr-like-enzyme-system-anthropic-says
- Malay Mail (via AFP), “Anthropic says its Claude AI helped uncover previously unknown enzyme system that may enable new gene editing.” https://www.malaymail.com/news/tech-gadgets/2026/09/24/anthropic-says-its-claude-ai-helped-uncover-previously-unknown-enzyme-system-that-may-enable-new-gene-editing/236311
Primary source
- Anthropic, “Claude discovers a novel enzyme system with CRISPR-like repeats.” https://www.anthropic.com/news/claude-discovers-novel-enzyme-system
Additional reporting
- Phys.org, “Anthropic touts AI-led biology discovery.” https://phys.org/news/2026-09-anthropic-touts-ai-biology-discovery.html
- IBTimes UK, “Claude Finds Hidden DNA System With CRISPR-Like Features That Could Cut, Copy and Paste Genetic Material.” https://www.ibtimes.co.uk/anthropic-claude-discovers-crispr-like-enzyme-system-1821665
Editor’s note on sourcing This finding rests entirely on Anthropic’s own disclosure and a preprint the company commissioned and released; no scientist outside the company’s chosen reviewer has yet assessed it publicly, and the preprint has not been peer-reviewed. Amodei’s claim about an independent Stanford finding could not be corroborated beyond his own social media statement and is reported here as an attributed claim, not a confirmed fact. The system’s biological function, and whether it is genuinely “programmable” in the way CRISPR is, remain unconfirmed and are treated in this article as open questions rather than established facts.
