Every message you send in Signal or Google Messages relies on the Signal protocol to keep it private. On November 10, 2026, Google DeepMind said its researchers used AI agents to mathematically prove key security properties of the actual code behind that protocol. The work took four weeks. Google says it has historically taken expert teams years.
The system is called AlphaProof Code. Human experts decide what needs proving. A team of AI agents, powered by Gemini Carbon, then writes the proofs in Lean, a computer language built for checking mathematics line by line.
Key facts
- Announced November 10, 2026 by Google DeepMind.
- Agents wrote millions of lines of machine checked proof in Lean in four weeks.
- The scope is Signal's core cryptographic code, not the whole Signal app.
What they proved
The agents proved security properties such as keeping messages protected before and after a device is temporarily compromised. Google says the proofs reach down to the code that runs on devices, not only the protocol on paper.
Google gives two figures for the size of the work: the page summary says over three million lines of proof and the article body says over six million. Either way, nobody writes that by hand in four weeks.
How they checked it
- Human experts reviewed every specification the agents worked from.
- The agents flagged about one percent of the codebase for a person to look at.
- To test the system, the team put back two bugs that had been fixed before. The system produced proofs that those buggy versions were insecure.
Signal's CTO, Ehren Kret, praised the approach, and Signal engineers gave feedback on the method.
What it means for a small business
You will not run AlphaProof Code yourself. The reason to care is the direction. Security checks that used to cost a fortune in expert time are getting cheaper because AI agents do the grinding work and humans check the result. Over time that should mean more of the software you depend on, from messaging apps to payment code, gets checked to a higher standard.
Today's takeaway is smaller and practical: Signal and Google Messages are the apps this work covered. If you discuss customer details or prices with staff over text, an end to end encrypted app is the safer default.
Our take
This is a research result, and Google says scaling it needs more research. The four week figure stands out, and so does the honest scope: Google verified the core cryptography, not every line of the app. Treat it as good news about the tools that protect your messages, not as a product to buy.
Frequently asked questions
What is AlphaProof Code?
AlphaProof Code is a Google DeepMind framework in which AI agents write formal proofs, checked in the Lean language, that a piece of software has certain security properties. Human experts choose what to prove and review the specifications.
Does this mean Signal has no bugs?
No. The work covered Signal's core cryptographic code and specific security properties, not the whole app.
Which AI model did the agents use?
Google says agents powered by Gemini Carbon wrote the proofs, with a mix of models used to translate security goals into formal specifications.



