{"slug": "openais-10000-agents-cracked-navier-stokes-in-88-hours", "title": "OpenAI’s 10,000 Agents Cracked Navier-Stokes in 88 Hours", "summary": "OpenAI deployed roughly 10,000 concurrent AI agents on September 1 that produced a 165-page proof of the Navier-Stokes existence and smoothness problem in 88 hours, a $1 million Millennium Prize question unsolved for over a century, with the proof formalized in Lean by GPT-6 Astra over 17 additional hours. NYU mathematician Tristan Buckmaster alleges OpenAI was aware of his parallel work with Levent Alpöge, adopted his approach, and pressured him to exclude Alpöge from credit, with an OpenAI researcher reportedly saying, \"Why would you ruin your career?\" OpenAI denies using their unpublished work, and the proof covers forced Navier-Stokes, not the unforced version specified for the Millennium Prize.", "body_md": "OpenAI deployed roughly 10,000 concurrent AI agents on September 1. Eighty-eight hours later, they had a 165-page proof of one of mathematics’ hardest open problems — the Navier-Stokes existence and smoothness problem, a $1 million Millennium Prize question that has resisted human mathematicians for over a century. The announcement landed today, September 8. So did the counterpunch: NYU mathematician Tristan Buckmaster alleges OpenAI was aware of his parallel work, adopted his approach, then pressured him to strip his Anthropic-employed co-author from credit. When he refused, an OpenAI researcher reportedly told him: *“Why would you ruin your career?”*\n\n## What OpenAI Actually Proved\n\nThe [Navier-Stokes equations](https://www.scientificamerican.com/article/openai-claims-blockbuster-math-breakthrough-amid-swirl-of-controversy/) describe how fluids move — water, air, blood. For over 150 years, a central question has gone unanswered: do these equations always produce well-behaved, smooth solutions, or can they “blow up” — generating physically impossible infinite velocities at a point in time? The Clay Mathematics Institute put a $1 million prize on the answer in 2000. OpenAI’s agents claim to have constructed a blowup: a scenario where the equations break down.\n\nThe technical scale is striking. Agents were split into parallel groups pursuing opposite hypotheses — some working toward smooth solutions, others toward blowup. Codex acted as a coordination layer, extracting insights from each group and cross-pollinating the most useful ones. The system exchanged 2.7 million messages, consumed roughly 130 billion output tokens, and cost OpenAI millions of dollars. The resulting proof was then formalized in Lean — a proof assistant that mechanically verifies logical consistency — by GPT-6 Astra, which spent 17 additional hours on verification. That Lean step matters: this isn’t just an AI’s confident assertion. The logic has been machine-checked.\n\nOne significant caveat is already circulating among mathematicians: [the proof covers *forced* Navier-Stokes](https://alphasignal.ai/news/openai-s-10-000-agent-swarm-solves-a-100-year-old-math-mystery) — a version where an external forcing term is applied to the equations. The Millennium Prize specifies the unforced version. The Clay Institute has not yet ruled on whether this satisfies prize criteria. Fields Medalist Terence Tao praised the underlying approach as “a remarkable achievement,” but has not formally endorsed the full claim. The distinction between forced and unforced Navier-Stokes is real and material. Headlines declaring OpenAI “solved” the problem deserve that friction.\n\n## The Credit Dispute That Overshadowed the Proof\n\nOn August 15 — two weeks before OpenAI launched its agent swarm — Buckmaster and his collaborator Levent Alpöge used a closely related blowup method to prove that forced Euler equations (a simpler fluid model) can produce singularities. The Euler result was a known steppingstone to Navier-Stokes. Buckmaster says the overlap is not a coincidence he is willing to dismiss.\n\nOn September 3, sensing that rumors were circulating, Buckmaster emailed OpenAI directly. Three days later, OpenAI’s Sebastien Bubeck called him. According to [Buckmaster’s account reported by TechCrunch](https://techcrunch.com/2026/09/08/openai-fought-dirty-on-career-making-math-problem-says-nyu-mathematician/), Bubeck offered two options: publish concurrently with OpenAI following Buckmaster’s work, or have Buckmaster solo-author a paper crediting an unnamed OpenAI model — but excluding Alpöge, because his Anthropic employment was “annoying.” Buckmaster rejected both. When he warned he would go public, Bubeck allegedly responded: “Why would you ruin your career?” and “If you don’t want me to be nice, then I don’t have to be nice.”\n\nOpenAI flatly denies using Buckmaster and Alpöge’s unpublished work. “We did not use their prompt or proofs to prompt our models,” the company said. Bubeck says he “came into the discussion following academic norms.” Both parties published on September 8 — [OpenAI’s Navier-Stokes proof](https://openai.com/index/navier-stokes-solution/) and Buckmaster’s three proofs alongside his full written account of the dispute. Alpöge, notably, works at Anthropic and used both Claude and Codex in his research with Buckmaster — a detail that makes OpenAI’s alleged demand to remove him all the more pointed.\n\n## What This Means If You Build With Agents\n\nThe capability shift here is real, regardless of who is right about credit. Until this week, AI agents primarily assisted human research — summarizing papers, writing code, pointing at prior work. This is different: 10,000 agents coordinating to produce original frontier mathematics in under four days. The cost remains prohibitive today (millions of dollars, 130 billion tokens), but AI compute costs have historically fallen faster than anyone predicted. The pattern — divide a hard problem across thousands of parallel agents, cross-pollinate insights, formalize the result — is a template that will be applied to drug discovery, materials science, and software verification well before this decade is out.\n\n## The Open Question Nobody Has Answered\n\nIn September 2025, Alpöge sent Buckmaster a private email: “I am worried about a future in which corporations, rather than individual mathematicians, end up holding claim to major pure-math results.” That was twelve months before it happened to him.\n\nThe Clay Institute has not ruled. Peer review of a 165-page AI-generated proof has no established process. No framework exists for attributing credit when an agent swarm does the work. OpenAI says it does not intend to claim the prize money — framing this as evidence of capability rather than a formal scientific result. That is a convenient position for a company that may have just outrun the academic process it depends on for credibility. For more on how [the credit controversy unfolded in full](https://www.axios.com/2026/09/08/openai-math-solution-navier-stokes-credit), Axios has the detailed timeline.\n\nThe math might be right. The ethics are not so clean.", "url": "https://wpnews.pro/news/openais-10000-agents-cracked-navier-stokes-in-88-hours", "canonical_source": "https://byteiota.com/openai-navier-stokes-agents-credit-dispute/", "published_at": "2026-09-08 19:12:18+00:00", "updated_at": "2026-09-08 19:18:01.719109+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "ai-agents", "ai-ethics"], "entities": ["OpenAI", "Tristan Buckmaster", "Levent Alpöge", "Sebastien Bubeck", "GPT-6 Astra", "Codex", "Clay Mathematics Institute", "Terence Tao"], "alternates": {"html": "https://wpnews.pro/news/openais-10000-agents-cracked-navier-stokes-in-88-hours", "markdown": "https://wpnews.pro/news/openais-10000-agents-cracked-navier-stokes-in-88-hours.md", "text": "https://wpnews.pro/news/openais-10000-agents-cracked-navier-stokes-in-88-hours.txt", "jsonld": "https://wpnews.pro/news/openais-10000-agents-cracked-navier-stokes-in-88-hours.jsonld"}}