Google DeepMind AI verified Signal's code in 4 weeks

Google DeepMind's new AlphaProof Code framework used Gemini Carbon agents to mathematically prove key security properties of Signal's real encryption code in 4 weeks, a job that used to take years. What it covers, and its limits.

Adam · News editor, Dear my…

· 3 min read

On this page

Short answer

On November 10, 2026, Google DeepMind said AI agents helped it mathematically prove key security properties of the real code behind the Signal protocol, the encryption that protects chats in apps like Signal and Google Messages. The work took 4 weeks, a job DeepMind says used to take expert teams years. It didn't find any new flaws, and it covers the core cryptography, not the whole Signal app.

What happened

DeepMind introduced AlphaProof Code, a framework where human experts and AI agents split the work the way a research team would. First, researchers worked with several AI models to turn Signal's security goals into formal mathematical specifications, and humans reviewed every one. Then agents powered by Gemini Carbon wrote the proofs in Lean, a language that lets a computer check each step of a proof automatically.

According to DeepMind's technical paper, the agents produced about 6.5 million lines of machine-checked proof code, resting on about 26,000 lines of trusted specifications. The agents also audited their own work and flagged roughly 1% of it for a human to look at. The proofs cover Signal's post-quantum PQXDH handshake and its Triple Ratchet, including forward secrecy and protection after a device is temporarily compromised.

The key detail: most past verification checked the design of a protocol. DeepMind says this is the first time the compiled code of a widely used protocol like this has been formally verified at this depth, by any method, with or without AI. To test the system, the team put two old, long-fixed bugs back into an isolated copy of Signal's code, and AlphaProof Code showed that the buggy version was not secure.

Signal's CTO Ehren Kret said in DeepMind's announcement that the work gives Signal "even greater confidence in our code's privacy guarantees." DeepMind is open-sourcing the specifications and proofs.

Why people are talking about it

Encrypted messaging is something billions of people use without thinking about it, and Wired notes that the same protocol also underpins WhatsApp. The announcement reached 600+ points on Hacker News, and Wired, Forbes and The Register all covered it on launch day. For many readers it was a rare AI story about making everyday tech safer.

What it means if you use AI

  • AI can check, not just create. We usually hear about AI writing code. Here it's proving that code does what it promises, with every step checked by a machine.
  • Your private chats get extra assurance. Nothing changes in your app, but the cryptography protecting your messages now has a much deeper independent check.
  • Know the limits. The Register points out the proof assumes the underlying cryptographic building blocks are perfect, and Wired notes the full mobile app wasn't verified.

Quick checklist

  1. Keep Signal, WhatsApp or Google Messages updated, so you run the code that gets checked.
  2. Remember that encryption protects messages in transit. A locked phone still matters.
  3. Turn on disappearing messages for chats you don't need to keep.
  4. Curious about the details? DeepMind's paper and Lean proofs are public.

Sources

  • Google DeepMind (Nov 10, 2026): Introducing AlphaProof Code: a breakthrough in verifying complex software at depth (primary)
  • Google DeepMind (Nov 10, 2026): AlphaProof Code: Verifying Signal's Security at the Machine-Code Level (technical paper)
  • Wired (Nov 10, 2026): A New AI Tool From Google DeepMind Proved the Signal Protocol Is Secure
  • The Register (Nov 10, 2026): Google DeepMind melds humans and AI agents to vet Signal's secure messaging techniques
  • Cover photo: "Young people texting on smartphones using thumbs" by Tomwsulcer, Wikimedia Commons, CC0

Sources read on November 11, 2026 (Europe/Prague).

Related reading:

Sources

  1. Google DeepMind (Nov 10, 2026): Introducing AlphaProof Code: a breakthrough in verifying complex software at depth (primary source)
  2. Google DeepMind (Nov 10, 2026): AlphaProof Code: Verifying Signal's Security at the Machine-Code Level (technical paper)
  3. Wired (Nov 10, 2026): A New AI Tool From Google DeepMind Proved the Signal Protocol Is Secure
  4. The Register (Nov 10, 2026): Google DeepMind melds humans and AI agents to vet Signal's secure messaging techniques
  5. Cover image: Young people texting on smartphones using thumbs, by Tomwsulcer, Wikimedia Commons, CC0

Written by our editorial team with the help of AI tools. Spotted a mistake? Tell us in the discussion.

Was this article helpful?

Discussion0 comments

Join the discussion

Log in or create a free account to comment. Only your nickname is shown.

No comments yet. Start the conversation.