Lean Language

Theorem prover

Leanstral: Open-Source foundation for trustworthy vibe-coding | Mistral AI

First open-source code agent for Lean 4.

https://mistral.ai/news/leanstral

Lean Programming Language

Lean is a theorem prover and programming language that enables correct, maintainable, and formally verified code.

https://lean-lang.org/

Seed-Prover

FormalQualBenchp

FormalQualBench

FormalQualBench: 23 graduate-level theorems, formalized in Lean4, with comparator-verified benchmark results.

https://www.math.inc/formalqualbench

수학, 인공지능, 그리고 형식화 [2]: 증명의 형식화와 인공지능

앞에서 우리는 인공지능을 어떻게 수학 연구에 활용할 수 있는지에 대해서 알아보았습니다. 이번 편에서는 최근에 떠오르고 있는 증명의 형식화formalization에 대해서 알아보고, AI, 특히 LLM을 활용한 수학 연구에서의 형식화의 역할과 형식화에 있어서의 AI의 역할과 주의할 점에 대해서 알아보면서 이 시리즈를 마치고자 합니다. AI가 증명을 했다고 주장하는데, 확실한가요? 앞서 언급했듯이 ChatGPT Pro나 Gemini Deep Think 등 최전선에 있는…

https://horizon.kias.re.kr/33348/

Palomar – a registry of Lean verified mathematics

In recent months there has been a proliferation of AI-generated proofs of various old and new results, some of which have been formalized in the proof assistant language Lean. However, checking tha…

https://terrytao.wordpress.com/2026/08/18/palomar-a-registry-of-lean-verified-mathematics/

Leonardo de Moura

Leonardo de Moura

Leonardo de Moura is a Brazilian computer scientist, and creator of the Z3 Theorem Prover and the Lean proof assistant during his time at Microsoft Research. He currently works at AWS and is the Chief Architect at the Lean FRO.

https://en.wikipedia.org/wiki/Leonardo_de_Moura

• Soundness bugs: bugs have been found in the Lean4 kernel that allow false propositions to be proved, raising concerns that AI could exploit such vulnerabilities.

• Mathematical consistency: the author argues that the consistency of the official Lean4 type theory has not yet been proven.

• Nested inductive types: complex recursive structures may be the main source of both theoretical and implementation-level vulnerabilities.

• con-leche: a separate verification kernel was built, but it does not guarantee the safety of the official Lean4 kernel.

• Anthropic's Fermat's Last Theorem: criticism of the decision to disable the sandbox while verifying AI-generated proofs. That said, the author notes that they re-verified it with the sandbox enabled and it passed.

A critical look at Lean4 — LessWrong

Lots of the discussion around Lean certificates and trust centers around general philosophical ideas of how much humans should trust formal certifica…

https://www.lesswrong.com/posts/zHhGa3iQBQcdSifFL/a-critical-look-at-lean4-1