Hacker News AI · 10/6/2026, 10:58:01 AM
Study Warns AI Autoformalisation Risks Semantic Loss in Navier-Stokes Proof
A new study argues that autoformalising natural language mathematical proofs, such as OpenAI's claimed Navier-Stokes blow-up proof, into formal languages like Lean risks semantic disconnection due to unresolvable ambiguities. The translation process is shown to sit at the highest level of the Solvability Complexity Index (SCI), theoretically harder than the Halting problem, implying current AI verification lacks reliability for complex mathematics.
SOURCE COVERAGEOriginal coverage
Abstract is increasingly used to verify mathematical texts, including those generated by AI, as in OpenAI's announced proof of blow-up of solutions to the Navier-Stokes equations. In this process, an AI system translates the text from a natural language (NL) into a formal language such as Lean. Once this translation is done, the argument expressed in the formal language can easily be mechanically verified. The purpose of this article is to demonstrate why this process may offer no confidence in the original NL argument, owing to the various difficulties in performing the translation semantically faithfully. In particular, we highlight that the problem of resolving ambiguities in mathematical NL text, which is necessary in order to provide semantically faithful translation, is arbitrarily high up in the Solvability Complexity Index (SCI) hierarchy/arithmetical hierarchy (the SCI =∞). Hence, informally, providing semantically faithful AI autoformalisation is harder than any computational problem including the Halting problem (which has SCI =1). To demonstrate the effect of this result we provide several examples of AI mistranslations of NL statements and proofs into Lean in practice, resulting in mismatches between NL proofs and their Lean `verifications'. These include OpenAI's announced Navier-Stokes proof. In particular, we show that the formalised Lean proof does not correspond to the NL proof of blow-up of solutions to the Navier-Stokes equations.
Comments pages, 4 Figures Subjects of PDEs (math.AP); Artificial Intelligence (cs.AI); Logic (math.LO) MSC classes, 03Dxx (primary) and 68V20, 68Txx, 03B65 (secondary) Cite as:arXiv.08144 [math.AP] (or arXiv.08144v1 [math.AP] for this version) https://doi.org/10.48550/arXiv.2610.08144
arXiv-issued DOI via DataCite (pending registration)
Submission history
From: Alexander Bastounis [view email] [v1] Tue, 6 Oct 2026 10:58 UTC (1,080 KB)