A major breakthrough in mathematical artificial intelligence, heralded by researchers at OpenAI, was proclaimed and then promptly debunked in a whirlwind of social media posts, revealing deep-seated tensions and communication issues within the competitive landscape of AI development. The incident began with a now-deleted post on the social media platform X from a senior OpenAI manager, who declared that the company's anticipated GPT-5 model had "found solutions to 10 (!) previously unsolved Erdős problems," describing them as challenges that had remained "open for decades." This assertion, quickly amplified by other researchers at the company, suggested a monumental leap for AI, implying that a generative model could independently produce novel proofs for difficult problems in number theory. Such a capability would signify that AI was not just a tool for processing existing knowledge, but a genuine engine for new scientific discovery, capable of unlocking solutions that have long eluded human mathematicians.

The excitement, however, was short-lived. The claim began to unravel almost as quickly as it had spread, largely due to the swift response from the mathematical community. The most crucial counterpoint came from Thomas Bloom, a mathematician who manages the website erdosproblems.com, the very source of the problems GPT-5 had supposedly solved. Bloom clarified that the term "open" on his site simply meant that he was personally unaware of a solution, not that the problems were officially unsolved by the global mathematics community. He characterized the researchers' statements as a "dramatic misinterpretation."[1] In reality, GPT-5 had not generated novel proofs but had instead successfully surfaced existing research papers and solutions that the website's maintainer had not yet incorporated. The AI had performed a powerful literature review, not a groundbreaking act of creation.

The public correction triggered a wave of criticism from prominent figures in the artificial intelligence field, highlighting a growing concern over the hype-driven communication strategies sometimes employed by leading labs. DeepMind CEO Demis Hassabis labeled the episode as "embarrassing," a pointed critique of the lack of due diligence from his chief competitor.[1] Yann LeCun, Meta's chief AI scientist, suggested that OpenAI had fallen victim to its own inflated expectations.[1] The incident fed into a perception that OpenAI, under immense pressure to demonstrate continuous, revolutionary progress, had become careless in its public announcements. The hasty claims and subsequent retraction raised serious questions about the internal verification processes for research findings before they are shared publicly, especially when those claims carry significant weight for the scientific community and the public's understanding of AI capabilities. The original posts were deleted, and the researchers admitted their error, but the affair left a lasting mark on the company's credibility.[1]

The episode underscores the immense pressure and high stakes involved in the race to achieve artificial general intelligence. The quest for AI systems that can master complex mathematical reasoning is seen as a critical benchmark for progress.[2][3] While this particular announcement proved to be a misstep, the field has seen genuine, significant advances. Ironically, around the same timeframe, OpenAI did achieve a legitimate milestone, with an experimental AI system earning a gold medal-level score at the prestigious International Mathematical Olympiad, a feat requiring deep, creative reasoning.[4][5][6][7][8] This genuine achievement stands in stark contrast to the Erdős problem debacle, illustrating that while AI's mathematical prowess is rapidly advancing, the communication surrounding these developments can be fraught with error and exaggeration. The incident serves as a cautionary tale for the AI industry about the importance of rigorous verification and responsible communication. As AI models become more powerful, the need for transparent and accurate representation of their capabilities is paramount to maintaining scientific integrity and public trust. The rush to declare the next big breakthrough can easily lead to embarrassing corrections that ultimately undermine the very progress researchers are working so hard to achieve.