[REVISION HISTORY]

OpenAI Navier-Stokes mathematical breakthrough

Updated 5 times since CLSTR started tracking revisions of this situation.

What changed

2026-09-10 23:04 UTC → 2026-09-10 23:59 UTC · added ~~removed~~

OpenAI announced a breakthrough in addressing the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. The company reported that approximately 10,000 autonomous AI agents worked for 88 hours, exchanging nearly 3 million messages and utilizing roughly 130 billion tokens to generate a proof involving a “finite-time blow-up for forced Navier–Stokes” under specific conditions. The solution, which addresses cases C and D of the Millennium Prize formulation, was subsequently formalized and verified in the Lean software proof assistant using GPT-6 Astra in 17 hours. The announcement has drawn controversy from the mathematical community. Researchers Tristan Buckmaster and Levent Alpöge have raised concerns that OpenAI may have been influenced by or gained access to their ongoing research into related fluid dynamics problems. Buckmaster suggested that the AI’s rapid progress on a “highly specific and unusual mathematical path” was unlikely to be a coincidence. OpenAI has denied using specific researcher data or private drafts, though the company acknowledged that general usage patterns from its products could have potentially aided model improvement. In response to criticisms, OpenAI modified its scientific article to cite mathematicians Diego Córdoba and Luis Martínez Zoroa. While the discovery has implications for meteorology, engineering, and the automation of formal ~~verification in fields like cryptocurrency security,~~ verification, the result has not yet been independently verified or officially recognized by the Clay Mathematics Institute. Mathematician Terence Tao has noted that while autonomous systems can generate complex solutions, there is a risk that “deep human understanding may be lost” if the process lacks sufficient transparency or human involvement. The event has ignited a broader debate regarding whether AI-generated answers provide the same conceptual value as human-led deductive reasoning.

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.