Close Menu
    Special Report

    OpenAI Has 100 Math Proofs in a Drawer. The Excuse Is the Mathematicians.

    superadminBy 10 Mins Read
    “Tired of the same tired news cycle? Escape to the full unfiltered feed at Whatfinger.com — America’s Frontline Truth Feed. 85%+ of readers come direct every day.” –  CLICK HERE for the FULL deal.🛑

    Story and MUCH info is below this top clip

    OpenAI told mathematicians it had the proofs. Then it sat on them. Anthropic is in the same race. The guild asked for a slower rollout. The labs treated that as cover. We at Whatfinger News aren’t falling for anyone’s BS. We’re calling it right here and now.

    OpenAI Has 100 Math Proofs in a Drawer. The Excuse Is the Mathematicians.

    Wired’s Isabella Ward published the fight on Tuesday. OpenAI is preparing to drop more than 100 solutions to unsolved problems, and the people who do this for a living are calling the labs mobsters. Nestor Guillen, a visiting math professor at NYU, told Wired there is “a perception of mobster behavior” from the leading AI companies. That line is the headline. The number underneath it is the story.

    OpenAI spokesperson Lindsay McCallum put the number on the record. Training on a new internal model started August 28. On top of a claimed solution to the Navier-Stokes Millennium Prize problem, “this model has now resolved more than 100 long-standing open problems across most areas of mathematics.” That is the company’s sentence, not a critic’s paraphrase. The proofs have not been sitting in a university archive. They have been sitting inside a lab that is lining up a blockbuster IPO, next to a rival lab doing the same thing.

    The hold. McCallum’s 100-plus, the 39 days from the August 28 training start to the October 6 dump plan, the uncertified Clay prize, and the 25 Fields Medalists who asked them to slow down.

    In August, OpenAI called about 40 mathematicians in to talk about what happens if the machines outrun the field. People in the room say the company hinted that the models had already cleared hundreds of long-standing problems, and that representatives promised not to release the solutions all at once. McCallum now says the company is “not aware of” that assurance. Northwestern’s Bryna Kra remembers the ask from the other side of the table. Do not put them in a blog post or a tweet, the way the firm had done with 10 problems earlier that month. Write papers, so working mathematicians can absorb the arguments and use them. “Apparently, that input was ignored,” she told Wired. People familiar with the plans say the GitHub dump of those August results is the Tuesday release.

    So the hold is real. The excuse is a mess.

    The record, one fake win included so the later claims don’t get a free pass.

    The mathematicians did ask for a delay. On September 11, 25 Fields Medalists, Terence Tao among them, signed a declaration titled “A Severe Misalignment of AI in Mathematics.” Their case is that a proof is not the same thing as a science. Mass-announcing solutions, they wrote, leaves no time for a proper writeup, for isolating the new idea, or for citing the humans who got close. A week earlier OpenAI had claimed Navier-Stokes, one of the seven Clay Millennium problems, each worth $1 million if the institute accepts the proof. The medalists want the pace cut so the field can still think. That is a guild protecting its method. It is also a fair request if the alternative is a tweet and a PDF nobody can check.

    The labs heard “don’t upset them” and used it. Results that McCallum dates to a model run that began August 28 were still unreleased in early October, while OpenAI stood up an advisory group at the Institute for Advanced Study and told the press it was listening. Kra says the listening stopped at the paper. Wired’s sources say the dump is happening anyway. Academics told Ward the field now feels like a playground for OpenAI and Anthropic while both shops prep the offerings. A proof is a terrible place to workshop investor sentiment.

    Claimed versus checked. Lean files are not a Clay stamp.

    Anthropic is not the innocent bystander in this. Levent Alpöge, a mathematician on Anthropic’s payroll, spent nearly a year with NYU’s Tristan Buckmaster on the fluid equations, using both companies’ tools on what Buckmaster says was a personal project. On September 8, OpenAI announced a Lean-checked Navier-Stokes argument, about 10,000 agents, 88 hours, and a compute bill TechCrunch put around $22.5 million at Astra rates. Buckmaster says he had told an OpenAI mathematician on September 3 that he and Alpöge were on a related track. Meeting notes seen by Wired have Sébastien Bubeck, who runs this work at OpenAI, worrying that “Levent must be talking to his leadership right now” and asking what stops Anthropic from throwing its own compute at a Millennium problem. The notes also have Bubeck telling Buckmaster, “If you don’t want me to be nice, then I don’t have to be nice.” Buckmaster says one of the offers was that he could publish first if Alpöge’s name came off, and that when he threatened to go to the press Bubeck asked why he would ruin his career. OpenAI has denied improperly pulling the pair’s data. It has not published a timeline that makes the race look clean.

    That is the part the “we didn’t want to hurt their feelings” line cannot cover. Two companies sat on claims, watched each other’s mathematicians, and argued about whose name goes on the trophy. The public got a blog post.

    The race bill. The $22.5 million is TechCrunch at Astra rates, not a receipt.

    One correction belongs in the piece, because OpenAI has cried breakthrough before. In October 2025, then-VP Kevin Weil posted that GPT-5 had solved 10 previously unsolved Erdős problems. Thomas Bloom, who runs the Erdős Problems site, called it a dramatic misrepresentation. The model had found papers Bloom had not seen. Demis Hassabis called the episode embarrassing. Yann LeCun enjoyed it in public. Weil took the post down. The May 20, 2026 unit-distance result was a different animal: an internal model disproved a 1946 Erdős conjecture, and Noga Alon, Melanie Wood, Bloom, and Tim Gowers put their names on companion remarks. Gowers wrote that if a human had submitted it to the Annals of Mathematics, he would have recommended acceptance. August 1 brought ten more Astra results with Lean certificates, including a non-sofic group that had stood since 1999. Anthropic’s Claude formalized Fermat’s Last Theorem in Lean in 11 days. That one adds nothing to the mathematics. Wiles already proved it. A machine-checked copy is an engineering win, not a solved mystery.

    Navier-Stokes is still a company claim with a formalization, not a Clay certificate. “Resolved” is McCallum’s verb until outside referees finish. Anyone treating all 100 as settled is repeating the October 2025 mistake.

    The assessment is simpler than the guild letter. If the Lean files check, the hold was the scandal. A solved problem that stays inside an IPO candidate for six weeks is not caution. It is inventory. Kra’s request for papers was the right ask, because a GitHub dump nobody can read is not progress either. OpenAI took the request, missed the papers, and kept the option to splash the number. Anthropic’s employee was in the middle of the credit fight, and Bubeck’s notes show a lab worried about the other lab’s compute, not about whether a turbulence proof reaches the people who build pumps and planes. The Fields Medalists want conceptual understanding preserved. Fine. Publish the proofs. Attach the prior work. Let Clay, Bloom, and the skeptics swing. A profession that asks a company to slow a verified result so the profession can keep its shape is negotiating status. A company that agrees, then dumps the lot for the ticker, was never protecting the science.

    — The Whatfinger News Team: Luke and Lisa, with Grok on edits and graphics

    Weekly Specials

    Election Links – Nov 3rd is just a few blinks away! 

    Top Features and Vids across all Whatfinger sites

    Our Projects…

    Get on over to Whatfinger.com – Whatfinger News does millions of pageviews a month for a reason…. Truth, Justice, the American way, plus news on the Left as well – all links from all sites gives you the true news.  – CLICK HERE 🛑

    Share.

    Related Posts

    Add A Comment
    Leave A Reply