OpenAI internal models helped find and fix Lean kernel soundness bugs
RexDouglass · x · 2026-08-24
The Lean team published a postmortem on using OpenAI internal models to hunt for kernel soundness bugs. With help from Daniel Selsam, the AI discovered new issues in the official Lean kernel and runtime, all fixed in v4.33.1. The document details how agents were instructed to use tools like comparator to detect adversarial proof exploits.
More from Safety
- Rebutting "surveillance is nothing new": AI dragnet ≠ old CCTV — AIandDesign · 2026-08-24
- RAND Corp releases white paper on AI Security Priorities with field-wide agenda — joshua_saxe · 2026-08-24
- Anthropic asks candidates if they'd accept stock going to zero for safety — zacharynado · 2026-08-24
- Interactive site explains SynthID text watermarking: algorithm and six costs — neil_chilson · 2026-08-24
- Interactive Explainer Breaks Down SynthID-Text Watermarking and Its Costs — neil_chilson · 2026-08-24
- "I gave a random AI product read/write access to my email and nothing bad happened" — aarthir · 2026-08-24