# Anthropic's Claude proves existing code correct and finds long-missed bugs

> Original title: Anthropic's Claude now writes code and proves it safe automatically

- Source: The Decoder
- Published: 2026-10-23T11:32:19.000Z
- Bad Influence: https://badinfluence.news/items/yg3dsnkgzq4se9z1v7mtzhjw0
- Original: https://the-decoder.com/anthropics-claude-now-writes-code-and-proves-it-safe-automatically/

## Summary

Anthropic used Claude to produce mathematical proofs that existing open-source code is correct, drawing on test programs from 123 libraries in Google's OSS-Fuzz project. Claude also found hundreds of bugs it was allowed to fix, including a logical error undetected for over a decade. The proofs hold only under strict conditions, including a specific processor architecture, single-threaded programs and inputs under 64 MB.

## Why it matters

The source shows how an AI model proved existing code correct and found bugs that years of testing missed, while stating the proofs' limits.