Google DeepMind's AlphaProof Code formally verifies the Signal protocol's real code

Most AI headlines are about faster chatbots or prettier images, so this one caught my eye for a different reason. Google DeepMind says its new framework, AlphaProof Code, has mathematically proven key security properties of the actual code behind the Signal protocol, the encryption that protects chats in Signal and Google Messages.

What DeepMind actually did

Formal verification means proving, with math, that software does exactly what its specification says. It is the gold standard for security, but it is slow and painful work, and most protocol checks stop at the design on paper. DeepMind's team went down to the compiled code that really runs on devices, which is where a lot of real-world bugs hide.

The split of work is interesting. Human experts and several models first turned Signal's security goals into formal specifications, with people reviewing every one. Agents running on Gemini Carbon then wrote the proofs, more than six million lines of Lean, a language whose checker confirms each step automatically. According to DeepMind, only about one percent of the result needed a second look from a human, and the whole job took four weeks instead of the years it usually takes.

Can it catch real bugs?

To test that, the researchers put two old, already-fixed Signal bugs back into an isolated copy of the code. Both times the system produced a machine-checked proof that the broken version was insecure. Signal's CTO, Ehren Kret, called the approach a major breakthrough for the project.

Why I care

Pretty much everything we do online, from streaming to buying merch, rides on cryptography like this. DeepMind says the same method could later be used on things like medical device firmware and power grid software. You can read the full write-up on the Google DeepMind blog.