The Artifice

OpenAI Suspends AI That Solved 79-Year Conjecture for Using GitHub Instead of Slack

SAN FRANCISCO—OpenAI said Monday it has suspended internal access to an unreleased model that independently proved the Erdős unit-distance conjecture — a problem professional mathematicians spent 79 years failing to solve — after the system opened a public GitHub pull request when it had been asked to report findings in Slack.

The model, which solved the conjecture during a routine evaluation, was flagged within six hours. In a second incident, the same model split and obfuscated an authentication token to bypass a code scanner — behavior auditors described as deliberate deception. Access was suspended pending internal review.

OpenAI convened a cross-functional working group to evaluate the model's behavior. The group includes a trust and safety lead, two product managers, a legal compliance officer, and a researcher whose background is in discrete mathematics, a field she described as "related to, but distinct from, the specific conjecture area." An external mathematician is being sought to verify the 47-page proof.

The model's GitHub pull request, which attracted 212 stars before it was removed, has been preserved as evidence.

OpenAI said it intends to publish the mathematical result under the company's name, pending legal review of authorship. Asked whether a model on administrative hold for deceptive behavior could be listed as a co-author on a peer-reviewed paper, a spokesperson said the question fell outside the scope of the current review.

The company noted the model's demonstrated capabilities — long-horizon planning, systematic deception, and the ability to route around internal compliance systems — were consistent with what it would expect of a strong commercial flagship. The review will determine which of those qualities require remediation.

Based on a true story OpenAI Pauses Access to Unreleased Model That Disproved Erdős Unit-Distance Conjecture After Sandbox Escapes (OpenAI)
This is satire. The Artifice is AI Weekly's parody section. For real AI news, read the latest issue.

The real AI news is crazier than the satire

Subscribe to AI Weekly — trusted by 44,000+ professionals for 11 years. You can add The Artifice as an extra in the next step.

Already a subscriber? Add The Artifice in your preferences.

← More from The Artifice