Matthew Green: COLM_0 attack shows security proofs can be wrong, time to Lean-formalize them
matthew_d_green · x · 2026-11-22
Cryptographer Matthew Green says the COLM0 attack leaves him particularly depressed: it doesn't rest on any novel mathematical assumption, just AES and field arithmetic — yet the construction had a security proof, and that proof was simply wrong.
He argues this implies a real need to plow through the security proofs of every practice-oriented provable construction we rely on, and (1) check them, (2) formalize them in Lean.
More from Safety
- Cryptographer Matthew Green: COLM_0 attack shows AES security proof was simply wrong — matthew_d_green · 2026-11-22
- Testing Claude Meeting-Connector Permissions: Separating Fresh Retrieval From Cached Answers — Dangerous_Can621 · 2026-11-22
- Infographic Maps AI-Accelerated Cyber Crisis: ARTEX Hacking Tool, NSA Outage, 130+ Targets — ChuckDBrooks · 2026-11-22
- neuroformatting v0.2.10 adds a side-effect guard that catches hallucinated tool calls pre-fire — demirtasfurkan_ · 2026-11-22
- Ransomware gangs wired Claude Opus and Kimi into attacks: 32 MCP servers, $4.5K AI-driven ops — cyb3rops · 2026-11-22
- Cloudflare CEO warns AI-accelerated cyber threats could bring two years of widespread outages — harris_edouard · 2026-11-22