Artificial Intelligence
OpenAI releases 700 AI manuscripts on open problems
OpenAI released more than 700 AI-generated manuscripts claiming solutions to open maths problems, and mathematicians responded with fascination, frustration and concern about the effects on their work and the field.
- On 6 October 2026, OpenAI released more than 700 manuscripts and said they solved hundreds of open maths problems.
- Proofs and Prompts gathered more than 100 researcher responses after the release.
- Peter Scholze urged patience and warned that similar systems might one day help break widely used encryption.
- Terence Tao said the papers might contain ideas worth using later, but he regretted the lack of people who could discuss and teach the work.
OpenAI put out more than 700 papers on 6 October 2026, saying they solved hundreds of open mathematical problems. The blog Proofs and Prompts then assembled more than 100 reactions from researchers.
Peter Scholze said mathematicians should treat the episode with patience and keep human understanding central. He also warned that systems with this level of capability might eventually find ways around common encryption methods. Terence Tao said the papers appeared to contain ideas that could matter after further digestion, but he objected to the lack of people who could discuss, present or teach the work.
Alvaro Lozano-Robledo called the date possibly the most important in mathematics so far, while saying OpenAI had only reached a weaker version of the Riemann Hypothesis. Ben Green said the release covered about three-quarters of the goals in a European grant he had received in June, and Roman Sauer said he recognised a technique he had developed.
Hugo Duminil-Copin said the announcement hit him with unexpected force, and Tristan Humbert said it disrupted his dissertation work and postdoctoral plans. Henry Wilton criticised the lack of named authors and failure-rate information, while Sam Hughes framed the episode as a loss of beauty.
Danny Calegari and Constantin Kogler described reasons for optimism, and Martin Bridson said mathematicians should not let AI labs set the agenda. The Association for Human Mathematics had already called for a boycott of OpenAI, and 25 Fields Medalists had issued a joint warning about a mismatch between AI industry goals and mathematics.
Named in this story
People
- Peter Scholze
- Fields Medalist who urged patience and warned about encryption risks
- Terence Tao
- Fields Medalist who described mixed feelings and called for solidarity
- Alvaro Lozano-Robledo
- mathematician who stressed the significance of the release and the weaker Riemann Hypothesis result
- Ben Green
- mathematician who said the release covered most of his grant goals
- Roman Sauer
- mathematician who recognised a technique in OpenAI's solutions
- Hugo Duminil-Copin
- 2022 Fields Medalist who said the announcement hit with unexpected force
- Tristan Humbert
- PhD student whose dissertation problem was among those claimed solved
- Henry Wilton
- mathematician who criticised missing author names and failure-rate details
- Sam Hughes
- mathematician who lamented the loss of beauty
- Danny Calegari
- mathematician who said his work was unaffected and cited a Claude-assisted project
- Constantin Kogler
- mathematician who had worked with Astra on a problem and posted a paper after the release
- Martin Bridson
- mathematician who warned against letting AI labs set the agenda
Companies
- OpenAI
- released more than 700 manuscripts claiming solutions
- Anthropic
- company behind Claude
Organisations
- Proofs and Prompts
- math blog that collected researcher responses
- Association for Human Mathematics
- group that called for a boycott of OpenAI
Products and systems
- Astra
- OpenAI model mentioned in earlier problem solutions
- Claude
- AI model that wrote code for a mathematical animation soundtrack
How the source tells it
The piece was highly charged and quote-led, mixing alarm, grief and outrage with a smaller thread of cautious optimism.
- alarm catastrophic predictions about encryption and digital infrastructure, reinforced by repeated descriptions of shock and paralysis
- grief loss language centred on a lost way of working, beauty and a path not taken
- outrage destructive metaphors such as being bulldozed, hit with TNT and treated as slop
- hype/novelty historic superlatives and a narrative of breakthroughs piling up into a defining moment
- optimism balanced reporting of ideas that might be fruitful later and a few researchers saying they remained curious or impressed