The latest rupture follows OpenAI’s earlier claim to have solved the Navier-Stokes problem, a Millennium Prize challenge. That announcement brought accusations that the company had raced ahead of academic norms and may have benefited from researchers’ work in progress—an allegation OpenAI denied. The episode left a field built on slow scrutiny wary of a well-funded outsider moving at Silicon Valley speed.¹

In September, the newly formed Advisory Group on Mathematics and Artificial Intelligence (AGMAI) urged AI labs to release findings promptly through established channels, disclose models, prompts and computing costs, and avoid turning research releases into marketing. Its warning was blunt: companies should “refrain from treating the release of mathematical results as marketing vehicles.”²

Then, on Tuesday, OpenAI released 722 manuscripts spanning 372 families of results, saying an internal frontier model had produced the work. The company said it had drawn on AGMAI’s advice; OpenAI president Greg Brockman framed the effort as moving “towards acceleration of scientific discovery and improving quality of life for everyone.”Greg Brockman (gdb)³

The scale impressed even skeptics. One account of the release said it demonstrated that AI’s advances are likely to spill into new specialist fields, much as coding tools already have. Yet Stephen Wolfram offered the counterweight: “You can make a trillion theorems easily. The problem is most of those theorems are not ones that anybody will care about.”⁴

Verification is now the immediate test. By Thursday, OpenAI had withdrawn solutions to at least three problems after apparent errors, while only some proofs had been formalized for machine checking.¹ Critics say the release still falls short of AGMAI’s standard of ensuring human understanding: only 10 of 719 manuscripts included model reasoning, and a reported 42% had not undergone formalization.⁵

For advocates, automated formal proof could free mathematicians to pursue new concepts. For critics, the hard work begins after the model’s answer arrives: turning a claimed proof into knowledge the wider discipline can trust, teach and extend.