Anthropic says its AI model Claude has found a protein system that looks like Crispr. The company claims Claude spotted an enzyme with properties “reminiscent of Crispr,” the Nobel Prize-winning gene-editing tool, after searching huge genomic databases for unusual reverse transcriptases.
Anthropic made the announcement on September 23, and it concerned an experiment with Claude agents. The company says that roughly 950 of those agents ran at the same time, and they discovered the interesting genetic sequences within 21.5 hours.
How Anthropic Found the System
Claude was instructed by Anthropic’s team to sift through vast genomic archives in pursuit of examples of reverse transcriptases, the subject of marker “interesting new examples”. These proteins work by copying RNA into DNA, which runs counter to the usual cellular sequence wherein DNA is converted into RNA. Reverse transcriptases serve a number of purposes across other species, such as placing segments of DNA within an organism’s genetic material.
Anthropic says Claude first found more than 200,000 possible reverse transcriptases. Of those thousands, the model later settled on an “unusual” reverse transcriptase family with a long stretch of repeated DNA that looks much like Crispr. That system has been named ART, standing for array-associated reverse transcriptases, by Anthropic. It sits inside jumbo phages, which are large viruses that target bacteria.
One AI agent wrote: “I can see by eye a tandem repeat array … that’s a Crispr-like … repeat array?!” The agent also acknowledged that this system could be a retron, as noted in the technical report. Crispr and retrons are both bacterial immune systems, though retrons are not the same multi-tool Crispr is.
Reactions From Gene Editing Experts
Fyodor Urnov, a gene-editing expert at the University of California, Berkeley and director for therapeutic R&D at its Innovative Genomics Institute, offered warm words. “I sincerely compliment Anthropic for telling the world about their discovery,” he said. IGI is collaborating with Anthropic but was not involved with the new discovery.
Not everyone was impressed. “The experiments are still in the queue. The PR is already live,” a professor at Stanford University who focuses on integrating AI into genome engineering research, says Le Cong.
Anthropic has set up a wet lab for drug discovery. The company admits it doesn’t yet understand what the system does. (“We don’t yet understand what this system does,” the company said in an X post, “but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA.”) One physical experiment was included in a technical report, which has not been peer-reviewed.
The Diamond On Santa Monica Beach
Cong framed the discovery as a search problem. “Let’s say we are on Santa Monica Beach and trying to scan through all the sand to find a diamond,” he says. “AI found this thing that looks very shiny, and then you have to go back to the lab to know—is it glass? Is it a diamond?”
It makes a difference which way you look at it. Claude found something that resembles Crispr. Whether the enzyme system can cut DNA has yet to be determined. The question of whether it can be used as a gene-editing tool, and if so, whether it proves to be a useful one, remains open.
What Claude Actually Did
Claude’s role was pattern recognition. Identifying new reverse transcriptases can take months of mining genome databases manually, so Shipman says Anthropic’s ability to do this in a day is impressive. But he cautions against attributing the discovery entirely to an AI model. “I think we have to be careful about saying that Claude autonomously discovered something, because there are scientists involved in the study,” he says.
Jason Gill, a microbiologist at Texas A&M University, and his colleagues first identified the same reverse transcriptase in a 2021 paper on jumbo phages. The fresh development comes from Claude, who found repeats surrounding the enzyme that suggested it could be part of a Crispr-like system.
“These models are good at finding patterns, better than a person staring at it with their eyeballs can,” Gill says. “A person could do all this but you have to already have a hypothesis in mind.” He adds that ART appears to have no “obvious relationship” to any known Crispr system, and it’s on Anthropic to prove that it actually has gene-editing activity.
“The novel thing is how they found it, not what it is.”
Retron-based gene-editing systems have already been developed by Shipman’s lab, which means that retrons can be employed in this fashion.
The Training Question
The scientific community cannot judge what Claude was trained on, which means the findings lack transparency at a time when major journals demand that researchers make their code freely available so that other scientists can reproduce their work.
An investigator who had devoted years to these very enzymes grew concerned that the model had learned from his research, after he made still-unpublished results available through the public version of Claude. (“Claude was,” Anthropic confirmed this to the New York Times, “not trained on any user transcripts, and our molecular biology team has no such access, either.”)
The Lab Test Ahead
It would still take human scientists running tests in the lab to prove the next big thing in gene editing, even if Claude identified it first. Anthropic CEO Dario Amodei mused about this on X in a post about the discovery, “Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equip.”
The outcome remains a mere pattern at present, not a proof. The distance to that future is considerable. Something resembling Crispr appeared to Claude. Confirmation by humans will be required.
What We Make Of It
What stands out about the announcement is the method, not the discovery itself. Claude located a rare item among countless possibilities with a speed that surpassed a human’s.
It’s understandable that people like Cong have doubts. The report hasn’t been reviewed by outside experts. Instead of finding the enzyme directly, the system’s gene-editing function remains unproven. And the pattern Claude found was already documented in a 2021 paper, meaning Claude didn’ simply found the surrounding repeats that sit around it.
The opacity of Claude’s training data is a real concern. A researcher’s suspicion and Anthropic’s denial raise questions about reproducibility and credit. The company’s denial to the New York Times is not a substitute for transparency.
Shipman’s point holds up: what matters is the approach, not the chemical itself. Claude excels at finding patterns, yet that skill cannot substitute for actual bench work.
The news from Anthropic serves as a prompt to remember that AI operates as a tool rather than a substitute for human decision-making.
Source material: “Anthropic Says It Discovered a Crispr-Like System. Now What?,” WIRED.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.