On Wednesday, Anthropic announced that its AI model, Claude, had discovered a new enzyme system with DNA repeats that look like CRISPR's. The company calls the system ART, short for array-associated reverse transcriptases. It's found mainly in viruses that infect bacteria. About 950 Claude agents searched gene databases for 21 hours to find it. Anthropic released the work as a preprint that hasn't been peer reviewed. The company says it doesn't yet know what ART does. Gene-editing stocks fell that day. On Sunday, Mario Rodríguez Mestre, a computational biologist at the University of Copenhagen, disputed the claim. He told The New York Times that his team had studied the same enzymes for four years and had shared its unpublished work with Claude. Anthropic answered with a statement the same day.

1. Claude Spotted Something Nobody Had Described (Anthropic, Feng Zhang)

Anthropic says its agents found the pattern on their own. Humans gave them only a starting prompt.

The agents did the spotting on their own, the company says. Anthropic's scientists wrote the first prompt and ran the lab work. Claude did the rest. Its blog quotes one agent reading raw DNA and writing, "I can see by eye a tandem repeat array." The agent then counted the repeats, searched the scientific literature for earlier descriptions, and filed a report for human review.

The company says nothing published describes ART. Anthropic's Sunday statement said, "We are not aware of any previously published work describing the ART system we recently found." It added that Claude "was also not trained on any user transcripts," and that its molecular biology team has no access to them either. The blog grants that earlier studies had already found the enzyme itself. It says Claude "appears to be the first to notice the system's defining features."

A CRISPR pioneer says the finding merits more study. Feng Zhang, a professor at MIT and the Broad Institute, reviewed the preprint on Anthropic's behalf. He called it "genuinely intriguing" and said it "merits further investigation." Anthropic says it shared the result early on purpose, to show what Claude can do.

2. But These Enzymes Were Already Known (Mario Rodríguez Mestre, Seth Shipman)

The Copenhagen researcher says his team found these enzymes first and shared its findings with Claude along the way.

He has studied these enzymes since 2022. Mestre first found them that year in giant viruses that infect bacteria. His team nicknamed them "jumbotrons." "So, for me, the most important question is not 'Were ARTs already known?' They were," he told the Times. His team hasn't published its findings yet.

He shared his unpublished work with Claude for years. For three years, Mestre's group used Anthropic's models to write code and draft papers. He shared his dissertation draft and a manuscript describing the enzymes. Claude's analysis drew on a public RNA dataset from 2022. Mestre says his team analyzed that same data a year ago and shared the results with Claude. The Times reviewed his Slack messages and other records, which "seem to show that the AI was extensively involved in his research." "My concern is that this information was used to train future versions of the models," he said.

Other researchers in the field knew about them too. Seth Shipman, a bioengineer at the Gladstone Institutes and UC San Francisco, said, "Many of us know about these systems." Mestre says he can't be sure Anthropic relied on his work. "I mean, coincidences happen," he said. "Everyone in the field is doing more or less the same things. But we were aware of these repeats more than one year ago." He's winding down his projects with Claude and moving to other models.

3. It's an Early Finding, Not a Breakthrough (Lucas Harrington, Steven Salzberg, Philip Kranzusch)

Other biologists said last week, before Mestre spoke up, that Anthropic oversold the finding.

Spotting a strange gene cluster is the easy part. Lucas Harrington, a co-founder of the gene-editing company Mammoth Biosciences, made this case in a post on X on Wednesday. "The hard part, and where the real discoveries come from, is figuring out what the system actually does," he wrote. He said early, incremental findings shouldn't carry "the framing of a major discovery." He also wrote that he's "genuinely rooting for all of the frontier labs" to get into biology. Dave Ricks, Eli Lilly's chair and CEO, endorsed the post.

Calling it the next CRISPR went too far. Steven Salzberg, who heads the Center for Computational Biology at Johns Hopkins University, told Bloomberg that most biologists don't "announce an unproven, preliminary finding as if it's the next Crispr." Philip Kranzusch, a biochemist at Harvard Medical School, called it "a really exciting beginning observation." He added, "It is not what we would describe as a breakthrough in biology."

Where This Lands

Anthropic says it knows of no published work describing ART and that Claude was not trained on user chats. Mestre says his team found these enzymes first and shared that work with Claude, and he's moving his research to other AI models. Harrington and Salzberg say Anthropic oversold an early finding. Harrington says the real discovery will come from working out what ART does, and Anthropic says that work is ongoing.

Sources