On September 8, 2026, Hacker News was rocked by two colliding front-page posts. One was an official triumph from OpenAI: the lab claimed that its latest internal models had cracked a 90-year-old open problem in mathematical fluid dynamics. The other came from Tristan Buckmaster, a mathematics professor at New York University. Buckmaster published an exhaustive statement detailing his team’s interactions with OpenAI, openly questioning whether the tech giant had front-run and co-opted his unpublished research breakthroughs.
At the center of the earthquake shaking both Silicon Valley and pure mathematics is the Navier-Stokes existence and smoothness problem. Designated by the Clay Mathematics Institute in 2000 as one of the seven Millennium Prize Problems, it carries a $1 million bounty. In simple terms, these partial differential equations govern how fluids and gases move—from turbulent air currents rolling off an airplane wing to deep ocean thermal currents. The core mathematical dilemma asks whether solutions to these equations can break down in finite time, developing infinite singularities where velocity or energy explodes beyond physical reality.
For nearly a century, generations of brilliant mathematicians hit a wall trying to resolve the blowup question. Cracking it demanded profound geometric intuition to expose elusive, hidden structures. In OpenAI’s telling, however, genius human intuition was supplanted by brute-force industrial compute and an automated proof pipeline.
10,000 Agents in an 88-Hour Sprint: How Compute Bulldozed a 90-Year-Old Problem
The computational balance sheet behind OpenAI’s breakthrough is staggering. To conquer the equations, the lab orchestrated roughly 10,000 concurrent AI agents. Over the course of 88 non-stop hours, this swarm operated as an interconnected mesh, exchanging 2.7 million messages to cross-examine intermediate reasoning and consuming roughly 130 billion output tokens. To ensure mathematical rigor, the system dedicated another 17 hours to verifying the final proof line-by-line in the Lean formal proof language.
Figure: Official formulation of the Navier-Stokes existence and smoothness problem. Source: Clay Mathematics Institute (claymath.org)
The architecture essentially recreated an industrial factory staffed by 10,000 tireless, obedient mathematics graduate students. Provided with a viable strategic direction, they exhausted centuries’ worth of scratchpad explorations in less than four days. Along the way, the system even branched off 100 agents to tackle the regularity problem for the unforced Euler equations, resolving it in 50 hours. In the end, the cluster delivered a 100-page formal manuscript.
Before this compute foundry, research timelines once measured in months collapsed into days. OpenAI’s models undoubtedly produced formidable mathematical derivations. But in Buckmaster’s telling, the foundry knew precisely where to aim its drills only because it possessed a leaked blueprint.
September 1 to September 8: Where Did the Unpublished Preprint Go?
The controversy hinges on an invisible starting line. Months earlier, Buckmaster had teamed up with Levent Alpöge, a mathematician and researcher at Anthropic. Working with LLM-assisted workflows, the duo sought to crack the Navier-Stokes problem by advancing earlier rough forcing frameworks into smooth forcing—the most formidable theoretical obstacle in the entire literature. On August 15, they achieved a breakthrough. By August 22, they formally verified their first AI-assisted proof in Lean. Buckmaster described the machine-generated steps as the most terrifying mathematics he had ever read.
Just as they approached the finish line, word began to leak.
| Key Milestone | OpenAI Official Post | Professor Buckmaster’s Statement |
|---|---|---|
| Late August | August 28: Began training new model | August 15: Critical breakthrough; August 22: First formal Lean verification |
| September 1–3 | September 1: Initiated campaign after hearing rumors | September 3: Rumors spread that progress had reached OpenAI |
| September 5–6 | September 5: Problem officially resolved | September 6: Two phone calls; OpenAI repeatedly shifts its timeline |
| Data Usage | Stated no access to specific user data | Draft preprints placed in Codex; OpenAI gave no definitive answer when pressed |
Buckmaster documented two tense phone calls with OpenAI representatives on September 6. During the discussions, the company’s timeline repeatedly shifted. Initially, OpenAI claimed the model had only been supplied with the raw problem statement. Later, they conceded that an entire internal task force had been mobilized with massive compute resources—and that the initial prompt had been dispatched days after hearing rumors of Buckmaster’s work.
When pressed on whether their models ingested unpublished drafts submitted through Codex, OpenAI offered an evasive denial. The company stated that no team member looked at specific user data, but conceded that they “cannot rule out de-identified data improved the model.” On Hacker News, the competing submissions triggered an intense voting clash: Buckmaster’s statement surged to 1,053 points, while OpenAI’s technical announcement plateaued at 1,010. The developer community was riveted not by the Navier-Stokes proof itself, but by the timeline anomalies and data provenance questions.
When Even Whispers Trigger a Compute Stampede
Fields Medalist Terence Tao captured the chilling implications of the episode:
“What is scarce and valuable now is identifying promising problems. Even rumors that someone is working on a problem will trigger massive AI-driven efforts to bulldoze it.”
For centuries, mathematicians flourished in an open culture of tea-room blackboard sketches, conference hall banter, and informal drafts passed between colleagues. These casual exchanges served as the lifeblood of creative discovery. But in an era where clusters can synthesize 100-page formal proofs in 88 hours, any leaked whisper invites an automated stampede to strip-mine the insight.
Figure: Schematic of vortex inward spiraling and axial stretching. Source: OpenAI
Once an unpublished manuscript touches a cloud-hosted copilot, or a viable proof strategy circulates among peers, it slips out of the creator’s custody. In an age of compute surplus, finding the viable path is orders of magnitude more precious than executing the mechanical derivations. Independent researchers are no longer competing with academic peers; they are racing against industrial clusters and automated theorem-proving farms.
”Why Ruin Your Career?”: The Asymmetric Power Play of Corporate AI
Perhaps the most troubling revelation in Buckmaster’s account is the lopsided power dynamic between a lone academic and a corporate tech empire. Buckmaster disclosed that OpenAI proposed two patronizing “compromises.” In Option 1, Buckmaster would publish the simpler Euler equation finding first, allowing OpenAI to publish the marquee Navier-Stokes solution the following day. In Option 2, Buckmaster could appear as the sole author on the Navier-Stokes paper—provided he completely excised Anthropic researcher Levent Alpöge from the credits.
OpenAI representatives stated bluntly that everything would be straightforward if Alpöge did not work for a direct commercial rival. When Buckmaster refused to betray his co-author, the tone turned hostile: “Why are you trying to ruin your career?” and “If you don’t want me to be nice, I don’t have to be nice.”
This aggressive posture laid bare the hubris that comes with monopolizing frontier compute. Mathematical truth is no longer an open human pursuit; it has been reduced to corporate leverage for flexing dominance and wounding market competitors.
When Academic Secrecy Collides with the Industrial Compute Rig
The Navier-Stokes dispute does not merely alter the history of fluid mechanics. Buckmaster described it as mathematics’ “Deep Blue moment.” It has punctured the foundational compact of modern academia. Months of grueling intellectual spadework can now be reverse-engineered and completed by an AI cluster over a single weekend. The temporal buffer that allowed human researchers to verify, refine, and claim credit for their ideas has evaporated.
In Hacker News comment threads, mathematicians half-jokingly suggested setting honeypots: publishing deliberate false rumors and bogus proof sketches to trick OpenAI into burning millions of dollars in compute trying to verify dead ends. Behind the gallows humor lies profound anxiety over compute hegemony.
The intricate mathematical lemmas in OpenAI’s Navier-Stokes preprint will soon fade from public discourse. What will endure is the chilling realization across global research institutions: active, unpublished research has become unhedged commodity ore. Armed with massive compute rigs, corporate AI can trace the faintest rumor and hollow out an entire intellectual deposit overnight. The open academic culture that nurtured centuries of scientific triumphs may have just met its endgame.
Reference Links:
- OpenAI Official Blog
- Buckmaster Public Statement (cims.nyu.edu)
- HN Discussion (item?id=49605915)
- HN Discussion (item?id=49613262)
