OpenAI says an unreleased model solved one of mathematics' seven Millennium Prize problems in 88 hours, and mathematicians are now accusing the company of scooping unpublished work.
Key Points:
- OpenAI said roughly 10,000 AI agents produced a proof that fluid equations can break down in finite time.
- An NYU professor says the company started work only after hearing about his unpublished results.
- The Clay Mathematics Institute has not accepted the proof and still lists the problem as open.
OpenAI Navier-Stokes Proof Details
The company posted its result on Sept. 8, saying an internal system it has not released proved that the Navier-Stokes equations, which describe how fluids move, can break down in finite time.
Roughly 10,000 autonomous agents ran the problem in parallel and reached an argument after 88 hours. The resulting manuscript runs past 160 pages.
That push started Sept. 1, after rumors reached OpenAI that two Millennium problems had fallen to other researchers. GPT-6 Astra then spent another 17 hours formalizing the proof in Lean, a language that verifies each logical step, and executives put the cost of the run in the millions of dollars.
Also Read: LAPTOP Crashes 99%, Hunter Biden Rejects Scam Claims And Blames Bots
Tristan Buckmaster Disputes Credit
Tristan Buckmaster, a mathematics professor at New York University, alleged that OpenAI moved on the problem only after word of his unpublished results reached the company. He had worked for nearly a year with Levent Alpöge, a mathematician employed by rival lab Anthropic.
Both men leaned on OpenAI's Codex assistant while assembling their own proofs, and the company can train on those sessions unless users opt out.
The pair chose a narrow route through the problem that almost nobody else was pursuing, which made the timing look suspicious to Buckmaster. He says OpenAI mathematician Sébastien Bubeck proposed a compromise that would strip Alpöge of credit, then asked why he would ruin his career.
OpenAI denies that its researchers or agents saw the pair's work before publication. It says no specific user data went into the effort. The company concedes that de-identified data drawn from product use may have improved its models, though it argues the two proofs differ significantly.
Clay Institute Withholds Verdict
Terence Tao of UCLA praised the results from Buckmaster and Alpöge, while warning that competitive AI runs risk turning mathematics into a quota game of little real value.
Abhishek Saha, a mathematics professor at Queen Mary University of London, described the episode as the sort of thing mathematicians generally avoid.
Martin Bridson, who leads the Clay Mathematics Institute, called the announcement exciting but signaled that review will take time. The body still lists the problem as unsolved. The institute picked seven problems in 2000 with a $1 million award attached to each, and only the Poincaré conjecture has fallen since, proved in 2003 by Grigori Perelman, who declined the money.
Read Next: Apple Prices Its First Foldable iPhone Duo At $1,999, And Samsung Should Worry
