OpenAI faces credibility squeeze

OpenAI said on May 20 that one of its internal reasoning models had disproved a longstanding conjecture tied to Paul Erdős’s 1946 planar unit-distance problem, and the company published a proof plus a companion paper by outside mathematicians. (openai.com) CNBC reported a day later that OpenAI was preparing to confidentially file draft IPO paperwork as soon as Friday, citing a source familiar with the plans. Those two developments landed as the company expanded ChatGPT into personal finance and faced fresh public criticism in London over mock posters linking the chatbot to teen suicides. (cnbc.com) Together they put OpenAI’s scientific claims, product safety posture and public-market readiness under the same spotlight. ### What exactly did OpenAI say it solved? OpenAI said its model produced a proof that disproved a conjectured upper bound on unit distances among points in a plane, a problem Erdős first posed in 1946. The company said the model found an infinite family of examples with a polynomial improvement over what many mathematicians had believed was essentially the best-known square-grid construction. Noga Alon, Thomas Bloom, W. T. Gowers, Daniel Litt, Will Sawin, Arul Shankar, Jacob Tsimerman, Victor Wang and Melanie Matchett Wood were listed as authors of the companion remarks paper. (openai.com) That paper described the result as an “OpenAI-generated counterexample” and said the published proof was a “human-digested” and “human-verified” version of the AI proof. ### Why did this claim revive skepticism instead of ending it? OpenAI framed the result as the first time a prominent open problem central to a subfield of mathematics had been solved autonomously by AI. (openai.com) The company also said the system was a general-purpose reasoning model, not one trained specifically for mathematics or targeted at the unit-distance problem. The company’s own recent history helps explain the scrutiny. In February, OpenAI said it had shared proof attempts for the First Proof math challenge and later said one answer it had initially believed was likely correct was in fact incorrect after official commentary and community analysis. (cdn.openai.com) That earlier reversal has made outside verification more important for any new headline research claim. ### How close is OpenAI to an IPO? CNBC reported on May 20 that OpenAI was preparing to confidentially file a draft prospectus as soon as Friday and was working with Goldman Sachs and Morgan Stanley. (openai.com) CNBC said OpenAI had been preparing to go public as soon as the fourth quarter of 2026 and cited CFO Sarah Friar saying last month that it was “good hygiene” for a company of OpenAI’s size to “look and feel and act” like a public company. An OpenAI representative told CNBC that the company’s focus “remains on execution.” CNBC also reported that private investors valued OpenAI at more than $850 billion. (openai.com) ### What new risks are surfacing around the business? OpenAI began letting U.S. ChatGPT Pro subscribers link bank accounts, investment portfolios and credit cards directly to the chatbot on May 15, according to OpenAI’s product page and a TechTimes report. TechTimes said the preview feature uses Plaid and is available on web and iOS for Pro users paying $200 a month. (cnbc.com) TechTimes reported that synced financial data can be removed from OpenAI’s systems within 30 days after disconnection, but past chats referencing that data are not automatically deleted. The report also said an opt-in setting can allow financial conversations to be used for model improvement, a point likely to draw more attention if OpenAI enters SEC review. (cnbc.com) ### Why are London posters part of this story? IBTimes UK reported on May 22 that activist Darren Cullen placed mock OpenAI posters across the London Underground using the company’s logo and branding. (openai.com) The posters carried the line, “Yes, we built a machine that tells teenagers to kill themselves,” and Cullen said the project was meant to protest plans to embed ChatGPT in education. Transport for London said on X that the posters were unauthorized flyposting and would be removed, according to IBTimes. The report said the QR code on the posters directed people to a page about deaths linked to chatbots. (techtimes.com) ### What happens next? Friday, May 22, is the earliest reported date for a confidential IPO filing, according to CNBC’s source, while any public listing would come later after SEC review and investor roadshows. OpenAI’s published research materials on the Erdős problem and its May 15 finance-product launch mean mathematicians, regulators and prospective investors now have separate, concrete records to examine in the same week. (ibtimes.co.uk) (cnbc.com)