{
  "id": 12901776,
  "title": "OpenAI mistranslated mathematics into code for its Navier-Stokes proof",
  "url": "https://urgent.news/2026/10/08/openai-mistranslated-mathematics-into-code-for-its-navier-stokes-proof",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-08T16:03:20.000Z",
  "source": {
    "name": "New Scientist",
    "slug": "new-scientist",
    "url": "https://www.newscientist.com/article/2592824-openai-mistranslated-mathematics-into-code-for-its-navier-stokes-proof/"
  },
  "original_language": "en",
  "account": "OpenAI has been accused of making a subtle error in its proofs of the Navier-Stokes problem, a team of mathematicians claims. The error, while not rendering the proofs incorrect, raises questions about the reliability of mathematical results generated by AI models. Anders Hansen from the University of Cambridge noted that the AI has to \"read\" the proofs produced by large language models, creating an additional burden on mathematicians. OpenAI announced on September 8 that it had found a solution to the Navier-Stokes problem, publishing the proof in two versions - one in natural language and another in Lean code. However, the two proofs don't match, according to Hansen and his team. This discrepancy occurs during the process of auto-formalisation, where the AI attempts to create a Lean proof that compiles, sometimes altering the logical argument in the process. The issue was identified by comparing the two proofs, with a specific focus on Lemma 8.6. Notably, an equation in the natural-language proof requires a certain value to be below m + 4, while the equivalent value in the Lean proof is required to be below m + 5. The latter is mathematically weaker, meaning it allows more possible answers for the variable. OpenAI acknowledged the mismatch but stated it doesn't invalidate either proof. The team from the University of Cambridge took about two weeks to identify the true discrepancy, while OpenAI's agents spent 88 hours generating the proofs.",
  "summary": "When OpenAI announced its surprise solution to the Navier-Stokes problem, it produced one proof for humans and one for computers – however, they don't match",
  "key_points": [
    "OpenAI's Navier-Stokes proof contains subtle mathematical error",
    "AI translation process alters logical argument in proof",
    "Discrepancy identified in Lemma 8.6 between natural language and Lean code proofs"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}