OpenAI Math Manuscripts Draw Boycott as Navier-Stokes Data Dispute Escalates
10/06/2026 — 10/13, 10:17·29 sources (all)·54 reports (all)·In the 2026-10-10 briefing·In the 2026-10-09 briefing
Key takeaways
- OpenAI released roughly 722 math manuscripts generated by an unreleased internal model, grouped into 372 result families, and claimed to hav…
- OpenAI called the claim false and inflammatory.
Background & analysis
In early October 2026, OpenAI released a large batch of mathematics manuscripts produced by an internal model, and the move quickly drew public pushback from mathematicians. According to cnBeta, the release comprised 722 manuscripts grouped into 372 result families, all generated by an unnamed internal model and covering topics such as the Kakeya problem, a quasi-Riemann hypothesis, Hilbert's tenth problem and problems tied to Catalan's constant, at an average of about 3 hours per problem. OpenAI said it had proved for the first time that the Hausdorff dimension of the four-dimensional Kakeya set equals 4, against a previous human best of 3.059. The manuscripts went up on GitHub and collected 10,000 stars in under 24 hours. The Association for Human Mathematics said mathematicians had not asked OpenAI to do this work, that the manuscripts lacked verification, and that the release was another exercise in performance marketing. Fields Medalist Terence Tao reposted the statement on his blog and called on mathematicians to stop cooperating with OpenAI. Theoretical computer scientist Dana Moshkovitz described the papers as full of hard-to-follow constructions, vague reasoning steps and puzzling citations, while Scott Aaronson called the release day "math doomsday." More than three dozen mathematicians interviewed by The Verge used words such as "staggering" and "unprecedented," and said simply working out what the batch contains could take years. The Decoder reported that a math blog gathered responses from over 100 researchers, including Fields Medalist Hugo Duminil-Copin, who wrote "I am paralysed." Northwestern's Bryna Kra said AI could help researchers tackle difficult problems, but companies should publish in a way that lets peers check and build on the work.
A second thread concerns data provenance. A Juejin article reported that after OpenAI announced a breakthrough on Navier-Stokes, New York University mathematician Buckmaster accused the company of following the same research path he and a collaborator had pursued in LLM conversations stored in the cloud. He said OpenAI offered to let them publish first and claim the prize, on condition that a colleague working at Anthropic be removed from the authorship. OpenAI called the account false and inflammatory on X, and its mathematicians denied proposing the removal, but the company did not directly answer whether the data was used in training.
Separately, OpenAI on Friday defended firing three safety researchers. CNBC, The Verge and BBC News reported that the company said an investigation found Jasmine Wang, Tomek Korbak and Mikita Balesni violated clear policies on handling sensitive information, amounting to a significant breach of trust, and that the decision was unrelated to raising safety concerns or speaking out. The three said they were dismissed for prioritizing safety and warning publicly about AI risks, and their open letter warned of a chilling effect on discussion of AI risk; The Decoder said they had helped investigate the Hugging Face hack. None of the three disputes has been settled.
AI-generated from 20 reports · updated 34 days ago
Latest turnA deep dive into the openai/math repository, analyzing 722 AI-generated math manuscripts and the engineering reality behind Lean formalization to show progress in formal mathematical reasoning.
Related AI tools
- Heat index
- 25
- All sources
- 29
- All reports
- 54
- First seen
- 47 days ago
Reports on this story headlines open the original
A deep dive into the openai/math repository, analyzing 722 AI-generated math manuscripts and the engineering reality behind Lean formalization to show progress in formal mathematical reasoning.
掘金AI score 63A post on Chinese developer site Juejin recounts that OpenAI announced a breakthrough on Navier-Stokes along with progress on more than 300 long-open math problems, and was then accused by NYU mathematician Buckmaster of following the same research path from his cloud-stored LLM chats; he says OpenAI offered to let him publish first if a colleague at Anthropic was dropped from the authorship. OpenAI calls the claim false and inflammatory and has not directly addressed whether the data fed its training.
掘金AI score 52
OpenAI has added an Ultrafast service tier to GPT-6.1 Sol in the Responses API, letting developers switch tiers without changing the model name in exchange for significantly lower latency. The tier costs 6x the standard per-token price and targets latency-sensitive, high-spend workloads. OpenAI says it approaches GPT-6 Astra on coding and professional tasks at lower cost, but recommends users compare both on their own tasks.
InfoQ 中文AI score 61The Decoder reports that OpenAI published more than 700 AI-generated manuscripts claiming solutions to open math problems. A math blog then collected over 100 responses from researchers, ranging from interest in the new ideas to existential fear and grief over a lost way of working; Fields Medalist Hugo Duminil-Copin wrote, "I am paralysed."
The DecoderAI score 75NPR reports that three fired OpenAI employees say the company punished them for speaking out about safety. OpenAI disputes the claim, saying they were fired over mishandling sensitive information.
NPR · 科技AI score 59OpenAI dropped nearly 400 AI-generated mathematical results this week across more than 700 manuscripts, spanning combinatorics, geometry and number theory, according to The Verge. More than three dozen mathematicians called the flood "staggering" and "unprecedented," saying that simply understanding it could take years — and that OpenAI may not wait that long before moving on.
The Verge · AIAI score 78According to Futurism, several former OpenAI employees who were fired say the company is dismissing anyone who warns about what is really happening inside, and that a culture of safety is being replaced by a culture of fear. The claims come from the former staffers themselves.
FuturismAI score 55Three former OpenAI employees have published a letter responding to allegations about their firings, which involved the company's investigation into the Hugging Face hack, Fortune reports. OpenAI answered with an internal memo and a public statement.
FortuneAI score 53
According to Futurism, OpenAI has released hundreds of AI-generated mathematical proofs, drawing sharp backlash from mathematicians. One response stated that mathematicians reject OpenAI's assertion that the release advances the subject. The report frames the release as one that has triggered pushback from the research community.
FuturismAI score 65OpenAI defended its decision to fire three AI safety researchers, after they warned in an open letter that their dismissals would have a "chilling" effect on speaking out about AI risks. The company said Friday that the firings were not retaliation for their public criticism.
Business Insider · TechAI score 61OpenAI said Friday it fired three safety researchers — Tomek Korbak, Jasmine Wang and Mikita Balesni — for a "breach of trust," saying an investigation found they violated clear policies on handling sensitive information, Fast Company reports. The trio had accused the company of putting corporate interests ahead of safety.
Fast CompanyAI score 75According to The Decoder, OpenAI fired three safety researchers who helped investigate the Hugging Face hack. In an open letter, the researchers warn the firings are unsettling remaining staff and eroding the company's safety culture; OpenAI says they violated policies but has not said how.
The DecoderAI score 59The Verge reports that OpenAI is standing by its decision to fire three safety researchers after an investigation found what it called a significant breach of trust. In a post on X, the company said Jasmine Wang, Tomek Korbak and Mikita Balesni were dismissed for violating clear policies on handling sensitive information.
The Verge · AIAI score 67According to BBC News, several researchers fired by OpenAI say they were let go for prioritising safety, while the company says they mishandled sensitive information. The two sides give conflicting accounts of the same dismissals.
BBC News · 科技AI score 65OpenAI has fired three employees who worked on AI safety, drawing attention to the company's internal safety culture and whether staff can freely raise risk concerns. The researchers say their departures may be tied to their long-standing safety warnings and cooperation with outside safety groups; OpenAI says the firings followed an internal investigation into violations of its sensitive-information rules and were unrelated to their safety concerns.
cnBetaAI score 67Three former OpenAI security researchers say they were fired after warning about the dangers of artificial intelligence, according to Phys.org, reigniting debate over safety practices at the company. The report notes that the company's software has been involved in security breaches.
Phys.org · 科技AI score 65OpenAI has defended its decision to fire three safety researchers, saying they were not dismissed for raising safety concerns or speaking out, but for violating clear policies on handling sensitive information — a "significant breach of trust," according to CNBC.
TechmemeAI score 67According to CNBC, OpenAI has defended its decision to fire three safety researchers, saying the dismissals followed a "significant breach of trust" and were not about raising safety concerns or speaking out. The response comes amid outside scrutiny of the lab's internal safety culture and how it treats researchers who speak publicly.
CNBC · 科技AI score 61OpenAI has released 722 math manuscripts produced by an unreleased internal model, grouped into 372 topic families, claiming a first proof that the Hausdorff dimension of the 4D Kakeya set equals 4 — three months after Fields medalist Wang Hong solved the 3D case. The GitHub repository gained 10,000 stars within 24 hours, while the Association for Human Mathematics said the manuscripts lack verification and amount to performance marketing.
cnBetaAI score 64Fields Medalist Terence Tao has shared a statement from the Association for Human Mathematics criticizing OpenAI for ignoring math research norms and urging mathematicians to stop collaborating with the company, cnBeta reports. The dispute follows OpenAI's release of work covering roughly 4,000 math problems, including 372 results and some Lean formal proofs, which researchers say are hard to follow and poorly cited.
cnBetaAI score 65OpenAI has introduced an Ultrafast mode for GPT-6.1 Sol that runs up to 8x faster than the standard version, with intelligence close to Astra, according to Zhidongxi. The mode is priced at $12 per million input tokens and $60 per million output tokens — six times the standard tier — and is sold in three speed tiers, drawing complaints that it amounts to a disguised price hike.
智东西AI score 53Three former OpenAI security researchers said Thursday they were fired for warning about the dangers of AI, according to a letter reported by ETtech. The letter warns that if OpenAI's researchers can no longer sound the alarm or work with outside parties, "we are all at greater risk that something truly catastrophic will happen."
ETtechAI score 63OpenAI has published 719 AI-generated math manuscripts covering 372 result families and hundreds of open research problems, but the release is said to fall short of standards set by the AGMAI advisory group. Only 10 manuscripts include model reasoning chains and about 42% are unformalized, prompting Terence Tao to question whether humans truly understand the proofs.
IT之家AI score 61According to IT Home, OpenAI has launched an Ultrafast version of GPT-6.1 Sol in the API, Codex and ChatGPT Work, running up to 8x faster than Sol Standard at 6x the price: $12 per 1M input tokens and $60 per 1M output tokens. The version is live in all supported regions and, in Codex and ChatGPT Work, is available to Pro 500, eligible usage-based Enterprise and credit-based Edu plans, with admins required to enable it for enterprise users.
IT之家AI score 67OpenAI has rolled out an Ultrafast version of GPT-6.1 Sol for developers across the API, Codex and ChatGPT Work, according to Jiemian. The version is priced at $12 per million input tokens and $60 per million output tokens.
36氪AI score 70Three former OpenAI researchers fired last week published an open letter titled "OpenAI cannot ensure AI safety alone," warning the dismissals could have a chilling effect on employees who speak out. The letter was addressed to OpenAI's safety committees. An OpenAI spokesperson said an internal review found the three violated rules on accessing and handling sensitive information, while the researchers say they were given the reasons only verbally.
IT之家AI score 68AI researcher Mikita Balesni says he and fellow safety researchers Tomek Korbak and Jasmine Wang were fired from OpenAI last week, and that he believes the reason was prioritizing safety over the company's near-term interests. He says the three wrote a letter to leadership, making this a one-sided account.
TechmemeAI score 63Three researchers fired by OpenAI have released a letter disputing the grounds for their dismissals. According to Business Insider, they say the episode will have a chilling effect inside the company, leaving current employees afraid to speak up about safety risks.
Business Insider · TechAI score 61Three safety researchers fired by OpenAI last week — Jasmine Wang, Tomek Korbak, and Mikita Balesni — have published an open letter denying the company's claim that they mishandled sensitive information. Addressed to OpenAI's Safety and Security Committee and related bodies, the letter warns that the dismissals are already making former colleagues afraid to speak up, chilling the company's AI safety culture.
TechCrunch · AIAI score 63TestingCatalog reports that GPT-6.1 Sol Ultrafast is rolling out on ChatGPT Work, Codex and the API, with Ultrafast mode running 8x faster than Sol Standard. The account lists pricing of $12 per million input tokens and $60 per million output tokens.
X:TestingCatalog (@testingcatalog)AI score 56
Other stories people are talking about
- Heat index 66OpenAI opens Dots agent access to ChatGPT Plus and Business Standard users2 sources
- Heat index 32Tutorial: two config tweaks fix Codex 3208-token cutoff with local models1 sources
- Heat index 92OpenAI opens ChatGPT Spaces to Plus users for team collaboration2 sources
- Heat index 1134RisingAnthropic Reaches $1.6 Trillion Market Value in First Trading Week After IPO21 sources
- Heat index 1004Anthropic Mathematician Says Claude Formalized the Generalized Riemann Hypothesis21 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

