RC RANDOM CHAOS

DeepMind's AI Formally Verifies Signal Encryption

· via Schneier on Security

Original source

Formal Verification of the Signal Protocol Using AI

Schneier on Security →

DeepMind researchers employed AI agents to establish formal mathematical proofs of the Signal protocol’s core cryptographic guarantees, completing the proof-of-concept in just four weeks. This work matters on two fronts: it provides independent validation of the encryption protecting millions of users across Signal, Google Messages, and WhatsApp, while also showcasing AI’s emerging capacity to solve complex mathematical problems at industrial scale.

The team translated Signal’s security objectives into formal mathematical specifications, with human experts reviewing each one. AI agents powered by Gemini Carbon then generated the proofs—spanning over six million lines of Lean code—and performed initial auditing. The agents flagged only about one percent of the codebase for human review, dramatically reducing manual effort while Lean’s automatic verification handled the remainder.

The verification uncovered no new vulnerabilities in the current protocol but correctly identified known flaws in older Signal versions, validating the method’s reliability. DeepMind is making the formal specifications and proofs publicly available to advance future verification efforts across the field.

Read the full article

Continue reading at Schneier on Security →

This is an AI-generated summary. Read the original for the full story.