A doctoral researcher at the University of Copenhagen says Anthropic’s biology announcement overlaps with his own unpublished work, and three other named scientists are challenging the claim from different directions. CNN’s Katie Hunt reported the pushback on Thursday. Anthropic has not answered it in anything CNN quoted.
The original claim, in brief: Anthropic said last month that AI agents working in its biology lab found a repeating DNA pattern beside the gene for an odd reverse transcriptase, an enzyme that copies RNA into DNA, in viruses that infect bacteria. The company called it a “novel enzyme system” resembling the bacterial machinery behind CRISPR gene editing. By Anthropic’s account, roughly 950 agents spent 21 hours picking out 3,500 of these enzymes, narrowed them to 20, and did it with “only high-level direction from our scientists.”
Aaron Engelhart, an associate professor of genetics, cell biology and development at the University of Minnesota, supplies the most useful caution. He calls the work “very interesting preliminary results” that “don’t demonstrate gene editing or provide a mechanistic picture of exactly what’s happening.” Here is the gap in plain terms. CRISPR works because bacteria keep an array of repeated DNA that stores snippets of past invaders, which guide an enzyme to cut matching DNA. Anthropic’s agents found a layout that looks like such an array. Looking like a gene editor is a hypothesis about a sequence on a screen. Editing genes means someone has to show the system cutting or rewriting DNA, and Anthropic’s own research document says its team has not yet worked out what the sequences do. The work is also not peer reviewed.
The priority question comes from Mario Rodríguez Mestre, a doctoral student at the University of Copenhagen who says Claude was a heavy part of his just-finished dissertation work. He says he put dissertation drafts and analyses on his private Claude account, and that they covered these viruses’ DNA patterns. Writing to CNN, he says his own unpublished study reports the identical discovery: “It is essentially the same finding.”
Mestre is careful about what he is not saying. “I am not claiming that Anthropic deliberately took our work. I cannot demonstrate that,” he told CNN. He lays out two possibilities. One is that information from his work somehow reached the model and was trained on. The other, which he currently finds more plausible, is that far more prior scientific knowledge and human direction lay behind the search than “autonomous discovery” suggests. Both are his hypotheses. Anthropic has conceded neither, and CNN says its emailed query to the company drew no reply. Our own source carries no statement from Anthropic on the allegation.
Ilya Finkelstein, a professor of molecular biosciences at the University of Texas at Austin, makes a separate point. Independent groups landing on the same result is not unusual in science, he noted, but an allegation like Mestre’s may make many researchers think twice before typing unpublished results into AI tools. That is an observation about incentives, not a verdict on who is right. Finkelstein also pointed out that four of the paper’s six authors are senior molecular biologists.
Yuval Elani, an associate professor of biochemical technologies at Imperial College London, argues something else again. Biology is not maths, he says, and until models are paired with robots that can do the fiddly experimental work, claims of AI discovery in biology are hype. Anthropic says it recognizes the limits of AI working alone and is investing in “model hardware” so agents can safely operate physical devices.
The episode exposes a structural problem. When one company runs the experiment, writes the account, and announces the result, the burden of proof falls on that company, and the reader cannot check the agents’ work. A priority dispute is harder still. A student’s dated drafts are one kind of evidence. What went into a model’s training data is another, and only the developer can inspect it. Settling this would take records from both sides: Mestre’s timestamps and Anthropic’s agent transcripts and prompts. Neither has been shown publicly.
Gene-editing stocks fell after the announcement, according to CNN. Until an outside lab tests whether the system actually cuts DNA, that move priced in a hypothesis.
CNN, reporting by Katie Hunt, published 8 October 2026.
