OpenAI prepares to dump hundreds of mathematical solutions on GitHub without papers

In August OpenAI gathered roughly forty mathematicians to discuss what happens if its models surpass human ability in the field. According to participants, company representatives hinted the models had already solved hundreds of long-standing open problems — but promised not to release everything at once, a promise spokesperson Lindsay McCallum now says the company is "not aware of." The mathematicians left with a mix of excitement and dread, calling the meeting a promising first step.

Bryna Kra, a mathematician at Northwestern University, says the group asked OpenAI to publish proper papers so the community could digest and use the results, rather than dumping them in a blog post or a tweet — as the company had done with ten problems earlier that month. That input was ignored. Now, according to people familiar with the plans, OpenAI intends to upload hundreds of the solutions discussed in that meeting to GitHub as early as next Tuesday.

The credibility of the move is shadowed by a separate episode. In an official statement McCallum said that on 28 August training began on a new internal model that allegedly solved the Navier-Stokes millennium problem plus more than a hundred other open problems across most branches of mathematics. Yet in September, after hearing "rumors" that competitors were closing in on the same millennium problem, OpenAI deployed thousands of agents to capture it first. Tristan Buckmaster, a professor at New York University, accused the company of hijacking work he had done with Levent Alpöge — an Anthropic employee — using OpenAI's own tools.

A recording of a conversation obtained by WIRED captures researcher Sébastien Bubeck suggesting Alpöge be excluded from any paper because it would "complicate things," a claim Bubeck denies in an earlier public statement. Buckmaster refused. In the same recording Bubeck voices concern over what Anthropic might do with its compute resources to claim its own millennium problem, and appears to threaten Buckmaster that going to the press would damage his career. Leading mathematicians now describe the field as a playground where OpenAI and Anthropic compete for headlines ahead of massive IPOs, abandoning standard scientific publication and attribution practices.