OpenAI drops Caltech math hackathon after mathematicians cry 'slop'
The friction between AI and its neighbors is getting legal teeth this midday: a lab retreats from academia's loudest critics, an infrastructure startup hires a lab's alumni for its IPO run, and a Chinese court puts a prison term on jailbreaking a model.
OpenAI has withdrawn its sponsorship of the Caltech Mathathon, the student-run hackathon for research-level mathematics, after an open letter from mathematicians accused AI companies of advancing a campaign of "slop mathematics." The event — 100 teams, $20,000 or more in AI credits per team, 40 hours on the Caltech campus chasing open problems — had signed up OpenAI, Anthropic, DARPA's expMath program, a16z, Y Combinator and Cognition as sponsors. The letter argued that "AI companies see research mathematics as an advertising opportunity," racing to announce AI-generated proofs of prominent conjectures in poorly communicated social-media drops that damage the careers of human mathematicians proving the same results. OpenAI said in a statement that it recognizes AI's progress in math is disruptive and wants to engage the math community on integrating the technology; it is so far the only sponsor to leave, and the organizers say the event goes on.
Our take: this is the cost of the prestige-extraction race, not a victory lap for anyone. Mathematics was supposed to be AI reasoning's clean, undeniable showcase — instead it has become the first field where the pushback is organized, credentialed and effective. The community has been building to this since June — we covered it in Terence Tao warns AI is killing math's open tradition, and the open letter echoes the same complaint: labs extracting results without the mentoring, attribution and care that keep a field alive.
Fidji Simo, the former OpenAI executive, is joining the board of AI infrastructure startup Nscale as it prepares a US IPO at a $14.6 billion private valuation. The Wall Street Journal reports she will continue as a part-time advisor to OpenAI while taking the board seat at the London-based GPU cloud, which reports roughly $103 billion in contracted revenue after a major Anthropic deal and raised $3.5 billion in pre-IPO financing with Nvidia at the table, targeting a listing this month. It is another data point in the revolving door between the labs and the compute layer they depend on — a lab's leadership talent joining the governance of a supplier weeks before that supplier faces public-market scrutiny.
We tracked the raise itself in Nscale lines up $3.5B in pre-IPO cash with Nvidia at the table. Watch how public-market investors treat lab-adjacent governance: convenient alignment or a conflict written into the cap table.
A Chinese court sentenced six people to up to nearly four years in prison for running a paid porn chatbot built on a jailbroken AI model, a landmark case treating prompt-level jailbreaks as a manufacturing crime. The Yuyao City People's Court in Zhejiang found Zheng, a former programmer at a major game studio, and five others guilty of producing obscene articles for profit: the team rewrote system prompts and planted role cards to bypass the base model's safety filters, generating more than 9,000 static and 5,000 dynamic obscene images for paying users. Sentences ran from three years and one month to three years and ten months plus fines, and the judgment is final. The presiding judge's line — technical neutrality does not mean technical innocence — lands alongside a September 7 Supreme People's Court opinion ordering harsh punishment for AI-assisted obscenity and fraud.
The significance is who is liable: not the anonymous user pasting prompts, but the service provider who weaponizes the jailbreak at scale. For Chinese generative-AI operators it draws a clear criminal red line; everywhere else it is a preview of the accountability arguments coming.
What to watch: whether Anthropic, a16z and Y Combinator stay on the Mathathon's sponsor list — and whether OpenAI's promised "engagement" with the math community produces anything concrete before the event runs.
Should labs withdraw from academic events under community pressure, or stay and take the criticism head-on? Tell us in the comments.
Sources: OfficeChai · Yahoo News · The Wall Street Journal · Crypto Briefing · Hua Shang Wang (华商网) · Qu Dong (驱动中国) · OECD.AI incident tracker
