Formal peer verification remains the gatekeeper of scientific claims even in the age of AI. Discuss with a recent example.

Peer verification is the process by which an independent scientific community, not the claimant, certifies a result. The September 2026 controversy over OpenAI's claimed proof concerning the Navier-Stokes equations — one of the seven Millennium Prize Problems listed by the Clay Mathematics Institute in 2000, each carrying a US $1 million award [1] — shows that even machine-generated, machine-checked results must pass this gate.

The claim and its limits

• OpenAI announced on 8 September 2026 that an unreleased internal model, run as roughly 10,000 concurrent agents over 88 hours, produced a finite-time "blow-up" proof, checked in the Lean proof-assistant language [2][3].
• The result covers only the forced 3D case, not the general problem; the Clay Mathematics Institute has not certified it, so the problem remains officially open [1][3].
• The roughly 100-page proof had not been reviewed by independent mathematicians when the claim was publicised [3].

Why verification still gatekeeps

• Formal ≠ verified: Lean checks internal logical consistency, not whether the theorem proved is the theorem the prize demands — a judgment only the community makes.
• Institutional filter: the Clay Institute's own rules require publication and sustained community acceptance, insulating science from press-release science [1].
• Credit and ethics: rival researchers Tristan Buckmaster (NYU) and Levent Alpöge alleged unfair conduct and possible use of private coding-platform session data, showing that attribution disputes are settled by scholarly process, not corporate announcement [2][3].

Wider implications

• Consent and ownership of data generated on AI platforms echo India's Digital Personal Data Protection Act, 2023, which grounds processing in purpose limitation and consent [4].
• Trust in AI-assisted discovery depends on reproducibility, open proofs and disclosure of methods.

AI has clearly become a powerful engine of mathematical discovery, but discovery and certification are distinct functions. The way forward lies in open publication of AI-generated proofs, transparent data-use norms and institutional protocols for AI co-authorship — so that speed of generation is matched by rigour of validation, keeping science self-correcting.