The 88 Hours That Broke Mathematics

88 Hours

September 8, 2026, Courant Institute of Mathematical Sciences, New York University. Tristan Buckmaster stood at the blackboard in his office, writing an equation with his left hand. Chalk dust fell on his cuff. He did not brush it off. The coffee on the desk had gone cold, and the rim of the cup had a chip. He took a sip, then set it down. He texted his wife: "Today might be a big day." Then he turned his phone over, screen down.

He had just posted three papers on the Navier-Stokes equations. One said that under certain conditions, the equations would blow up. It was a step toward the Millennium Prize Problem. He had spent years on it. He had not issued a press release or contacted the media. He put the papers on the preprint server, then went to teach two classes. Between classes he stood at the end of the hallway, looking out the window, fingers tapping the railing. A student passed and asked him a question. After answering, he added: "Go look at the one I posted yesterday." The student said okay. Buckmaster nodded and walked back to his office.

OpenAI posted a tweet that same day. Their internal model had organized ten thousand AI agents and, in 88 hours, completed a proof of the existence and smoothness problem for the Navier-Stokes equations. Computing costs ran into the millions. Below the tweet was an image: a summary of the model's output, dense, thousands of lines. Likes passed ten thousand within half an hour.

When Buckmaster saw the tweet, he was taking antacid. He shook out two tablets and swallowed them with cold coffee. He stared at the screen for a long time, then opened his own paper page and refreshed. No change. He refreshed again. Still no change. He closed the browser, stood up, and walked to the blackboard. He looked at the equation he had written. He picked up the eraser, erased a symbol, wrote it back. Erased it, wrote it back. Finally he put the eraser down. Chalk dust stuck to his fingers. He went to wash his hands. The water was very cold.

WIRED published an interview the next day. Steven Strogatz, Cornell professor of mathematics, 67, science writer. The reporter asked how he felt about the recent AI breakthroughs in mathematics. He was silent for a moment. Then his voice began to shake. He said: "I'm really terrified." He took off his glasses and pinched the bridge of his nose with his thumb and forefinger. On the desk was a pothos, leaves a little yellow, a pair of scissors resting by the pot. He later posted on X: "The article didn't explain why I was moved. I'm not entirely sure myself." Other posts followed. "Mathematics has a 4,000-year history," he wrote. "To be part of that long conversation is an honor." He added that he enjoyed playing with AI and found it exciting. Then: "But science is thrilling, and the human wickedness that comes with it is also great."

He mentioned Tristan Buckmaster. Buckmaster reminded him of many mathematicians: pure, rigorous, loyal to truth, perhaps unprepared for how the corporate world does things. Strogatz called this tradition Wimbledon. White clothing, etiquette, a care for how the game is played. The US Open was different. Louder, more showy, more open, coarser. He said: "Yes, I love the US Open. I also love Wimbledon."

On September 11, Terence Tao, Yu Deng, and 25 other Fields Medalists released a joint statement. Titled "The Severe Misalignment of Artificial Intelligence in Mathematics," it argued that AI companies treat problem-solving as a benchmark to push forward. That harms mathematics as a science and the mathematical community. In an interview, Tao made a metaphor. He said AI companies "dump the carcass of raw meat on the table in our public village, say 'Here, I've solved your food problem,' and then walk away." As he typed, the edge of his keyboard cover curled up. He stopped, went to the kitchen, and poured a glass of water. The faucet was turned on too hard; water splashed onto the counter. He wiped it with a rag, folded the rag, and draped it over the edge of the sink.

The document continued: mathematical discovery requires five stages. Creation, verification, explanation, acceptance, and incorporation into textbooks. AI companies do only the first two. The rest is left to the mathematical community. Tao's own project used AI for formal verification. He found that the generation end had been accelerated, but the verification end still ran at human speed. Proofs written by AI were several hundred lines longer than human ones, redundant, lacking global restructuring. It could execute a restructuring plan specified by a human, but it would not discover restructuring opportunities on its own.

On July 23, Jacob Tsimerman announced at the Fields Medal press conference that he was joining OpenAI and turning to AI safety research. He said: "AI is at a point where, if you're going to get on the train, now may be the time, because things are moving very fast." There was a small ink stain on his shirt cuff. His wife sat in the audience, clutching the program, knuckles white. He had co-authored a paper on scenarios in which AI leads to human extinction. He did not say mathematics had no future. He said AI safety was more urgent than pure mathematics.

In the mathematics department hallway, someone asked: "Have you learned Lean?" Another said: "I'm grinding LeetCode." They laughed. The laughter was short. A doctoral student turned a recruitment flyer over and wrote a definition on the back. He wrote half, crossed it out, wrote again. Finally he folded the flyer into a paper airplane and threw it into the trash can. The paper airplane hit the rim and fell to the floor. He did not pick it up.

Cornell mathematics department lounge. The coffee machine was broken. Several flyers from quantitative trading firms were scattered on the table: starting salary 450,000 and up, a million not rare. A visiting scholar picked one up, looked at it for a long time, then put it back. He picked up his cup, went to the pantry, rinsed it with hot water, and did not add coffee. He took a sip of hot water and stood by the window. It was raining outside. He stood for ten minutes, then returned to his desk and opened a paper. He read the first paragraph, then read the first paragraph again. He picked up a pen and drew a question mark in the margin.

At the end of the interview, Strogatz said: "In the future, if you want to do breakthrough mathematics, you won't be able to compete without AI." He paused, then added: "I also love Wimbledon." He turned off the recorder. It was raining outside. He picked up his umbrella and found that one of the ribs was broken. He put it back, put on his hat, and walked out into the rain.

Buckmaster later questioned OpenAI on Twitter, saying it had scooped his results. He said he had previously exchanged research ideas privately with Levent Alpöge, a researcher at Anthropic. The discussion on Twitter quickly turned into "Are mathematicians just sour?" He turned off his phone and went downstairs to the vending machine to buy water. The machine swallowed his coin. He pressed refund. Nothing. He patted the machine. Still nothing. He turned and walked away without buying water.

That night, he returned to his office. The equation was still on the blackboard. He looked at it for a while, picked up the eraser, and erased it all. Chalk dust floated in the air. He opened the window; the sound of rain came in. He sat down, opened a new file, and began to write. He wrote a word, deleted it. Wrote another word, deleted it. Finally he wrote a line and did not delete it. He stared at the line until the screen dimmed. He moved the mouse; the screen lit up. The line was still there. He saved the file. The file name was the date. He turned off the computer, picked up his coat. In the coat pocket was a flyer from a quantitative trading firm; he did not know when it had been stuffed there. He took it out, glanced at it, and put it back in the pocket. He walked out of the office and closed the door. The hallway light was voice-activated. He took a step; the light came on. He took another step; the light went out.