Google DeepMind says AI framework formally verified key Signal protocol code
Reported by Wired, Mezha, Startup Fortune and 7 more outlets10 sources ↓
Google DeepMind says AlphaProof Code verified security properties in compiled code for Signal’s cryptographic protocol in a four-week project. The analysis did not cover the entire app and found no new vulnerabilities in the code examined.
Key points
- The work covered compiled code for Signal’s PQXDH handshake and Triple Ratchet
- AI agents generated more than six million lines of Lean proof code; about 1% was flagged for human review
- The system identified two previously fixed bugs when researchers reintroduced them
- The researchers are releasing the specification and proof artifacts
- The analysis did not cover the full Signal app and found no new flaws in the portions examined
Sources
- WiredRead ↗A New AI Tool From Google DeepMind Proved the Signal Protocol Is Secure
- MezhaRead ↗Google DeepMind Used AlphaProof Code to Analyze the Signal Protocol
- Startup FortuneRead ↗Google DeepMind's AI proves key Signal protocol guarantees in four weeks
- Google DeepMindRead ↗Introducing AlphaProof Code: a breakthrough in verifying complex software at depth
- The Chenab TimesRead ↗Google DeepMind Unveils AlphaProof Code, Revolutionizing Software Verification with AI
Details
Google DeepMind used AlphaProof Code to check whether compiled code implementing key parts of the Signal protocol met specified security properties. The researchers focused on the PQXDH handshake and Triple Ratchet, including properties such as forward secrecy and authentication. The same protocol is used by Signal, Google Messages and WhatsApp.
The project took four weeks and involved a five-person team, according to finance.biggo. DeepMind says AI agents generated more than six million lines of Lean proof code, with about 1% of the codebase flagged for human review. The work built on earlier formal-verification work by the Signal Foundation.
Researchers reintroduced two previously fixed bugs into an isolated version of the code; the framework produced proofs showing the altered versions were insecure. DeepMind’s analysis found no new vulnerabilities in the code it examined. The researchers are releasing the specification and proof artifacts to support further work.
The verification did not cover the entire Signal mobile app, and its conclusions rely on specified security properties and assumptions about cryptography.
Written from coverage as of
Images
- Google DeepMind
- Wired
- The Decoder
- finance.biggo
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh