Anthropic Says Claude Discovered a CRISPR-Like Enzyme System in Bacteriophage DNA at Its New Biology Lab

Overview

Anthropic said its Claude AI model autonomously discovered a previously uncharacterized enzyme system in bacteriophage DNA with structural properties reminiscent of CRISPR, according to a company announcement introducing a new life sciences research group and wet lab. The company said the system, which it calls array-associated reverse transcriptases (ART), was flagged by Claude agents “with only high-level direction from our scientists,” as Anthropic put it.

What We Know

Anthropic formed the research group in the spring of 2026 to test “whether general AI models can systematize and accelerate” biological discoveries, according to Anthropic. The lab is located in the Bay Area and, according to the company, “looks like a typical molecular biology lab,” restricted to the lower biosafety risk levels, BSL-1 and BSL-2, and does not handle pathogens that can infect humans, as Anthropic described it. “All of the lab work is performed by human scientists,” the company said, and Claude has not been given autonomous control of lab equipment, a point also noted by TechCrunch, which reported that “all physical experiments are conducted by human scientists.”

According to Anthropic, the company gave Claude a prompt to search a database of DNA sequences for new examples of reverse transcriptases (RTs) — enzymes that copy RNA into DNA. Claude agents “gathered over 200,000 RTs, picked out 3,500 new candidate systems, and narrowed those to the 20 most compelling candidates that they analyzed to produce human-readable reports,” Anthropic said, adding that this kind of analysis “can take weeks to months” for an expert human scientist. The search that led to the discovery took “21 hours spent searching this data by roughly 950 agents using 210 million tokens,” per Anthropic — figures also reported independently by TechCrunch, which wrote that “the AI processed approximately 210 million tokens using roughly 950 agents to search through biological data.”

During that search, one agent examining an unusual RT family noticed a pattern in the surrounding DNA. “While combing through the raw DNA sequence near the RT, the agent exclaimed: ‘[The DNA next to the RT] is spectacular: I can see by eye a tandem repeat array … that’s a CRISPR-like … repeat array?!’” Anthropic wrote in its announcement. The agent then, in Anthropic’s words, “counted the repeats and measured their spacing, compared the layout with the known RT systems, and searched the literature for any previous report of the pattern” before filing a report for human review, according to Anthropic.

After laboratory testing, Anthropic’s scientists concluded the pattern marked a new enzyme system, which they named array-associated reverse transcriptases, or ART. Per the company, ART “is found mainly in bacteriophages and consists of three parts: the RT, a partner gene beside it, and a long array of evenly spaced DNA repeat sequences,” a layout that “resembles a CRISPR array,” according to Anthropic. The underlying RT itself, found in a jumbo phage, had been identified in prior research, but Anthropic 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 of unknown function.” Early lab experiments show the ART array “is also expressed as a set of distinct short RNAs,” Anthropic said, “suggesting that something analogous may be at play for this system” as with CRISPR’s programmable RNA-guide mechanism.

Feng Zhang, described by Anthropic as “one of the pioneers of CRISPR genome editing and a professor at MIT and the Broad Institute,” reviewed the company’s pre-print and said: “This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation. I hope this work encourages more scientists to explore how AI can support their research,” according to Anthropic.

Anthropic CEO Dario Amodei credited outside researchers too. He wrote on X that a team from Stanford had previously discovered a system “that is in some ways similar to the one Claude found,” as TechCrunch reported. Amodei also addressed the question of whether Claude might eventually run lab experiments itself, saying “eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren’t doing that today,” according to TechCrunch.

The new lab builds on other AI-for-lab-work efforts at Anthropic. The company noted it has “experimented with using AI to accelerate lab work with initiatives like the Model Hardware Standard, but this approach is less conducive to the sort of ad hoc workflows that are involved in our molecular biology research,” as Anthropic explained.

What We Don’t Know

Anthropic said its “work to understand the primary function of ARTs is ongoing” and that it does not yet know what the system does. “Further experiments are underway to determine how ART works,” the company said, framing the disclosure as an early result rather than a completed characterization, according to Anthropic. Feng Zhang’s comment that the finding “merits further investigation” points in the same direction — the system’s function and any potential applications remain unconfirmed pending peer review of the pre-print.

Analysis

The episode is notable less for the biology itself, which Anthropic frames as preliminary, than for the workflow: thousands of AI agents combing a public DNA database, generating and self-critiquing hypotheses, and surfacing a single anomaly that a human research team judged worth testing in the lab. Anthropic said it is now studying its own hypothesis-generation process, asking “what distinguishes the proposals we judge worth testing from those we set aside” so that the answer can be fed back into how Claude is instructed, according to Anthropic. Both Amodei’s and Zhang’s comments — one flagging Claude’s connection to a similar prior discovery at Stanford, the other calling for further scientific validation — signal that Anthropic wants the finding read as an early proof of concept for AI-assisted genome mining rather than a finished result.