An OpenAI rumor set the deadline: researchers rushed a Unique Games paper

Rumors of an OpenAI proof of the unique games conjecture pushed Dor Minzer and his students to publish a 95-page result in just three days.

On September 11, 2026, Dor Minzer, a professor at MIT and a leading expert on the unique games conjecture, began receiving messages asking whether he was close to settling one of the most famous open questions in theoretical computer science. The rumor: OpenAI, fresh from announcing a proof about the behavior of fluids, had allegedly discovered a proof of the conjecture and might publish it any day. As Quanta Magazine reports, the pressure that followed was real, and it arrived with a deadline nobody had chosen.

The conjecture is not an exotic curiosity. Subhash Khot posed it in a 2002 paper written while he was still a graduate student. In plain terms, it claims that some constraint satisfaction problems are brutally hard even when you will happily settle for a poor approximation. Picture a graph whose nodes must be colored, with every edge carrying a rule about the colors at its two ends. The conjecture says finding a coloring that satisfies 99% of the constraints can be as hard as it looks, and that relaxing the target to 1% or even 0.001% does not make the problem meaningfully easier.

The paper went online with a note that the manuscript was mathematically complete but not in the shape the authors wanted to share.

Khot's insight was counterintuitive in a specific way. The difficulty of such a coloring problem depends on the layout of the graph, the number of allowed colors, and the exact set of constraints. The claim is that in the hardest configurations, wanting a merely decent coloring buys you nothing. That is a strange and powerful statement, and it is why the conjecture has resisted proof for more than two decades while reshaping how the field thinks about approximation.

That is why a proof would ripple far beyond one puzzle. It would automatically imply that current methods for solving many seemingly unrelated problems cannot be improved, edging the field closer to a unified theory of computational difficulty. The questions researchers keep asking are ultimately about where the boundary between easy and hard sits, and this conjecture stakes out a lot of that territory at once.

Minzer had not proved the conjecture. He and his graduate students Yumou Fei and Shuo Wang had recently proved a milestone result about a closely related question, and they were in the middle of writing it up, a process that typically takes months. With the rumor hardening, they rushed. Three days after the first messages, they posted a 95-page paper online, prioritizing completeness over clarity, with a disclaimer on the first page stating that the manuscript was mathematically complete but not in the shape they wished to share.

The rush was vindicated, then overshadowed. On October 6, OpenAI announced a proof of the unique games conjecture along with 376 other results across many fields of mathematics. Researchers greeted the deluge with the now-familiar mixture of alarm and curiosity. Their reaction to Minzer's result was less ambivalent: Ryan O'Donnell, a theoretical computer scientist at Carnegie Mellon, called it another truly great one.

We find the timing more revealing than the trophy. A rumor out of an AI lab effectively set the deadline for a human result, and a team chose to ship a mathematically complete draft before it was polished. That is not a story about machines replacing mathematicians. It is a story about how AI announcements are already shaping the tempo of pure research, and about how much of that tempo stays invisible from the outside.

Opinion · Open question

OpenAI announced a proof of the unique games conjecture, and the announcement is not the same thing as a settled theorem. The story leaves one question hanging: how does a field digest 376 results at once, and who decides when a claim like this starts counting as knowledge? Until that plays out, the rushed milestone paper may be the more durable contribution of the week.

Related articles from LaRebelión

Wikimedia says unapproved OpenAI agents edited wikis and hammered its APIsOpenAI safety lead quits: the problem is culture, not rulesOpenAI Under Fire: California Subpoenas Rogue Hacking Agents

Original source: Quanta Magazine

Produced with AI support and reviewed by the newsroom

Byline · Chief editor · English edition · London

“A rumor from an AI lab set the deadline for a human proof, and that is the part worth remembering.”