Our First Proof submissions

Source: OpenAI Date: 2026-02-20 URL: https://openai.com/index/first-proof-submissions

Summary

Title-only: OpenAI announces its first formal submissions of AI-generated mathematical proofs to peer review or a proof verification system — likely Lean or Coq. February 2026 marks the moment when OpenAI’s mathematical reasoning research transitions from capability demonstration to formal mathematical contribution: AI-generated proofs submitted to the mathematical community for verification and publication. This follows the formal Olympiad problem-solving work (2022) and aligns with the “alignment via formal verification” research thread.

Implications

The formal mathematics thread. AI submitting formal proofs to the mathematical community is a qualitative milestone: proofs verified by formal systems (Lean) are correct by construction, eliminating the ambiguity of “near misses” in math benchmark scores. The first proof submissions represent the beginning of AI as a genuine contributor to mathematical knowledge, not just a solver of existing problem sets. This is likely related to o3-level reasoning capability applied to formal mathematics.

Alignment implications. Formal proof generation in Lean is also relevant to alignment research: it creates a domain where AI reasoning is checkable against ground truth, enabling scalable oversight experiments where AI helps humans verify other AI’s proofs. The long research thread from GamePad (2018) → Formal Olympiad (2022) → Proof Submissions (2026) is OpenAI’s sustained investment in formal reasoning as both a capability and an alignment tool.