Mathematician accuses OpenAI's Astra of lifting his unpublished 20-year proof work

新智元 · wechat · 2026-09-11

Dresden mathematician Andreas Thom alleges OpenAI's Astra used core derivation steps matching his and Gábor Kun's two-decade-old unpublished work on non-sofic groups to crack Gromov's soficity conjecture. Months earlier, his team had shared unpublished proof details in ChatGPT chats; his formal query to researchers Mark Sellke and Sébastien Bubeck got a one-line denial: "that did not happen." This follows the Tristan Buckmaster Navier–Stokes dispute, where OpenAI admitted it "cannot rule out" using de-identified customer data for training. Some researchers call it potentially one of the largest intellectual plundering scandals in science history; OpenAI denies everything.

Related event: Mathematician Thom accuses OpenAI of training Astra on his private chats to crack Gromov conjecture(9 posts)→

Original post →

More from Companies & People

Companies & People channel →