New Delhi: Weeks after OpenAI solved a Millennium Prize Problem in mathematics, Anthropic’s Claude has claimed a new scientific discovery in biology. In a statement published last week, Anthropic said Claude discovered a novel enzyme with CRISPR-like properties that could lead to revolutionary life sciences technologies.
However, a New York Times report on Sunday revealed that Mario Rodríguez Mestre, a biologist at the University of Copenhagen, said that his team had already discovered the enzyme a year ago and had used Claude in their work. They expressed worry that Claude used their data to reach its conclusions.
“Claude discovered a novel enzyme system with properties reminiscent of CRISPR, with only high-level direction from our scientists,” said the Anthropic statement, published on 23 September.
According to the statement, Anthropic’s scientists asked Claude to parse through a massive DNA database, looking for examples of reverse transcriptase (RT) enzymes. While usually, DNA makes RNA, which then makes protein, RTs are different. They are enzymes that can use an RNA template to make DNA and are used by scientists in gene editing, especially along with CRISPR in Prime Editing.
AI for biology
In early 2026, Anthropic developed its own biology research group with a team of scientists to advance Claude’s usage in life sciences. The programme, employed by Anthropic, was massive – Claude spent 21 hours analysing this data, using 210 million tokens across 950 AI agents.
It then found that along with some RT genes in the database, there were repeating patterns of DNA sequences. Moreover, these patterns were followed by an unidentified protein, and the scientists are still uncertain what it does. This repeat DNA system near the RT genes was named Array-associated Reverse Transcriptase (ART) by the Anthropic team.
Yet, this repeated DNA pattern along certain RT genes was an interesting discovery because this is also the exact pattern in which CRISPR – a powerful DNA modification technology – was identified.
“Its precise function, biotechnological utility (if any), or level of significance is not yet clear, but at minimum it is work I would have been proud to do as a PhD student,” said Dario Amodei, founder of Anthropic, on X. “We believe AI for biology is on an exponential thread.”
The scientists also released a preprint based on this discovery. The paper explained how, usually, a gene or enzyme followed by a repeating sequence often points to a programmable targeting system in DNA sequences. For example, in CRISPR, the repeated DNA is copied into small pieces of RNA called guide RNAs. These guides help the CRISPR enzyme find the matching DNA sequence, like a search instruction, and then cut or modify it.
While the exact function of ART is not yet known and the team is continuing to work on it, Anthropic’s statement said that it was important to share these updates early to demonstrate how Claude can help contribute to biological discoveries.
Mestre, whose team at Copenhagen had been working on the ART system and related enzymes for close to 4 years, said they had not published any of their findings yet. However, they had regularly used Claude to write code and draft manuscripts, sharing information about their research with it in the process.
The NYT article, meanwhile, said that Anthropic denied any training on user transcripts and was not aware of any published data on the ART system, implying that Claude made the discovery by itself.
However, this is not the first time that a closed-source AI model has been accused of ‘stealing’ a scientist’s work. Earlier this month too, when OpenAI announced that their agents had provided a solution to the Navier-Stokes equation, NYU Professor Tristan Buckmaster said that he had been working on a similar solution. He said that he was sceptical that OpenAI accessed his work, and it was used to build the prompts for their models.
(Edited by Aakriti Handa)