Skip to content

DeepMind's AlphaProof Code verifies Signal down to compiled code

Google DeepMind used AI agents to write six million lines of machine-checked proof in four weeks, formally verifying security properties of the Signal protocol at the level of compiled code, then open-sourced the artifacts.

Source: Google DeepMind

Google DeepMind announced AlphaProof Code on 10 November, an agentic framework that combines human expertise with AI agents to formally verify software. Its first target was Signal’s encryption protocol: the agents proved that the compiled code, not just the design on paper, fulfils key security properties such as keeping messages protected before and after a device is temporarily compromised, down to the exact code running on devices.

The agents, powered by Gemini Carbon, generated over six million lines of machine-checked Lean proof and flagged only one percent of it for a human to double-check. DeepMind says the whole process took four weeks, a task it describes as historically taking expert teams years. The verification artifacts are now open-sourced for anyone to build on.

Why it matters: Signal’s protocol also underpins Google Messages’ RCS encryption, so this is proof work on a protocol protecting billions of conversations, not a toy example. Formal verification catches whole classes of bugs that testing can’t, but it has stayed rare because it’s slow and expensive. If AI agents can genuinely cut that to weeks, it changes which teams can afford to prove security-critical code correct instead of just testing it.

Share

Related reading

More from The Wire

All briefs