Scientists say AI companies are scooping their research

At least twice in the past five weeks, researchers have said they were working for an extended period on a particular question when an AI company either answered it or announced results before them, according to Nature.

On September 7, NYU mathematician Tristan Buckmaster posted a paper with Anthropic's Levent Alpöge on a simpler version of the Navier–Stokes problem, a long-standing open question in fluid dynamics that the pair had worked on for a year as a personal project. The following day, OpenAI announced that its agents had solved it. The company said in a statement that an investigation had confirmed that Buckmaster's "prompts" in the two months before the 8 September announcement "could not have influenced the system in any way, including through training", and that OpenAI's researchers and agents "did not see any of their [Buckmaster and Alpöge's] work through any means until they released it publicly".

A similar episode landed at Anthropic, which announced that its agents had found certain viruses carry repeated DNA segments resembling CRISPR gene-editing systems. That is the same pattern Mario Rodríguez Mestre, a PhD student in computational biology at the University of Copenhagen, had been studying for years. Anthropic told the Times that its model was "not trained on any user transcripts".

"They're going to get scooped because the AIs are very good at solving problems, and they're going to get better and better," said Geoffrey Irving, chief scientist at Resolution, an AI-safety research organization in Berkeley, California.

Scientists are already changing habits. Sandra Laurentino, a reproductive epigeneticist at the University of Münster, now uses AI only to check why her code fails, and before doing so renames parameters to generic labels such as "group A has feature X"; the recent controversy, she said, "made me even more paranoid". Samuel Mehr, an auditory cognitive scientist at the University of Auckland, has forbidden his students from uploading protected information to commercial LLMs: "I think it's incredibly risky to hand out your intellectual property to third parties when you don't know what they're going to do with it."