AlphaProof Code: DeepMind agents prove Signal's security with 6.5M lines of Lean
LuizaJarovsky · x · 2026-11-11
Google DeepMind introduces AlphaProof Code, a multi-LLM agentic framework that achieved a long-standing grand challenge: formally verifying the actual code of the Signal protocol, protecting billions of private conversations.
Key facts:
- Human experts and AI first translated Signal's security goals into formal specs (each human-reviewed); agents powered by Gemini Carbon then produced the proofs, generating over 6 million lines of Lean against 26,000 lines of trusted specifications and AArch64 semantics.
- The proofs cover Signal's post-quantum PQXDH handshake and Triple Ratchet, establishing forward secrecy, authentication, post-compromise security, and handshake secrecy — the first formal verification of a widely deployed cryptographic implementation by any method, with or without AI.
- Formal verification was historically prohibitively expensive (seL4 took 20 person-years); the authors argue AI can make it practical for real-world systems. The Lean development is open-sourced.
More from Safety
- AI agents are finding critical CVEs — design your SaaS assuming the server is lost — haltakov · 2026-11-11
- Google Colab bans user for 'abusive content' over NSFW AI image generation — EdgyUsername_0529 · 2026-11-11
- Computational Linguistics journal submission system hacked, login credentials stolen — fredahshi · 2026-11-11
- Microsoft lobbies Trump admin to keep Chinese hardware in US AI data centers — kevinnbass · 2026-11-11
- Fable 5.2 reportedly lacks ZDR without a special deal, as safety classifiers rely on retained data — shaunralston · 2026-11-11
- Position paper: pluralistic alignment is thriving as research but absent from deployed frontier models — benno_krojer · 2026-11-11