Illustration representing Google AI agents auditing code and encryption libraries with digital verification proofs.
Crypto

Google AI agents audit Signal library code with millions of Lean proofs

Representative photo ยท Photo: Coolcaesar at English Wikipedia / Wikimedia Commons, CC BY-SA 3.0

By: NexVani DeskDate: 17 November 2026Views: 0

Google DeepMind researchers used AI agents to generate millions of Lean proofs connecting security guarantees to a compiled build of Signal's library.

This is news, not investment advice. NexVani Media is not a SEBI-registered investment adviser and does not recommend buying or selling any security. Do your own research or consult a registered adviser before investing.

Researchers at Google DeepMind revealed that their artificial intelligence agents successfully created mathematical proofs linking specific security guarantees to a compiled build of Signal's encryption library. The team published the resulting evidence online for public inspection following a report detailing roughly 6.5 million lines of machine-checked Lean proof code. The findings focus on a particular build of libsignal, which is the implementation of the Signal protocol, operating under explicit environment and cryptography assumptions.

Five people worked on the proofs over a four-week period, directing the work, choosing reference models, and reviewing the specifications that defined the tasks for the agents. The primary target was an AArch64 Linux executable running libsignal-protocol v0.105.4 alongside SPQR v1.6.0 within a command-line harness. The resulting theorems encompass session initialization as well as message sending and receiving, though key generation remains outside the verified scope. The Google report connects the compiled executable to properties of PQXDH and Triple Ratchet one-to-one messaging, addressing forward secrecy, authentication, and post-compromise security.

Two distinct proof models offer different levels of assurance. The symbolic analysis depends on idealized cryptographic operations and collision-free executions, whereas the computational analysis establishes bounds on classical attacker success under stated assumptions. However, the symbolic proof only guarantees handshake secrecy under specified later breaks of Diffie-Hellman, and the computational proof excludes quantum attackers. Furthermore, group messaging and multi-device state remain outside the scope of the report.

Google AI agents generated millions of Lean proofs to examine a compiled build of the Signal encryption library.

Understanding the google signal proof models and updates

A project update merged on GitHub verified ten additional library routines, reducing the number of trusted routine hooks from 37 to 27. Instead of relying on modeled contracts, the newer pinned executable verifies machine code for those specific routines. Additionally, the researchers reintroduced two previously fixed libsignal bugs to test whether the specification successfully rejected the known regressions.

Why it matters

Public Lean proofs expose both the software guarantees and the underlying assumptions of cryptographic implementations. This transparency gives researchers a concrete method to verify the claimed connection between security requirements and executable code.

Add NexVani Media as a preferred source on Google

Sources

ND

NexVani Desk

NexVani Media brings you the most important world news as it happens, with every story linked to its sources.