---
format: "aidr-story-markdown/v1"
id: "9e91233a64bf583079ce74273f9e2adc1a0784217f67f7fdb7c61f06f4f1cc2b"
canonical_url: "https://aidr.today/9e91233a?lang=en"
title: "Andrej Rushti Proves Transformer Invariants From Scratch Using Lean"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-16T05:47:23.000Z"
category: "Research"
topics: ["lean","transformer","formal-verification","theorem-prover","open-source","reasoning"]
source_urls: ["https://huggingnews.com/ai/andrej-rushti-proves-transformer-invariants-from-scratch-using-lean-1472f797","https://x.com/srush_nlp/status/2099972195988513169","https://marketbrief.now/ai/andrej-rushti-proves-transformer-invariants-from-scratch-using-lean-1472f797"]
summary: "A framework called Lean Verified Transformers uses the Lean theorem prover to provide formal mathematical proofs for the internal properties of AI models, establishing invariants for the Transformer architecture without external assumptions. Andrej Rushti's work, published on September 15, treats AI model internals as provable mathematical objects and explores the technical difficulty of applying formal verification to existing software code."
---

# Andrej Rushti Proves Transformer Invariants From Scratch Using Lean

> [Open the canonical story](<https://aidr.today/9e91233a?lang=en>)

**Published:** 2026-09-16T05:47:23.000Z
**Category:** Research
**Topics:** lean, transformer, formal\-verification, theorem\-prover, open\-source, reasoning

## Summary

A framework called Lean Verified Transformers uses the Lean theorem prover to provide formal mathematical proofs for the internal properties of AI models, establishing invariants for the Transformer architecture without external assumptions\. Andrej Rushti's work, published on September 15, treats AI model internals as provable mathematical objects and explores the technical difficulty of applying formal verification to existing software code\.

## Sources

- [Story source](<https://huggingnews.com/ai/andrej-rushti-proves-transformer-invariants-from-scratch-using-lean-1472f797>)
- [Story source](<https://x.com/srush_nlp/status/2099972195988513169>)
- [Story source](<https://marketbrief.now/ai/andrej-rushti-proves-transformer-invariants-from-scratch-using-lean-1472f797>)

