For years, proving that a widely used messaging app’s encryption code is truly secure was a job for teams of specialists working for years. Google DeepMind says its AI agents have now done it in four weeks.
The company announced on 10 November that a new system called AlphaProof Code had formally verified key security properties of the actual code behind the Signal protocol, the encryption technology that protects billions of private conversations in Signal and Google Messages. DeepMind describes it as the first time a widely deployed cryptographic protocol has been formally verified at this depth, and the first time the implementation of such a protocol, rather than just its design, has been formally verified by any method.
Formal verification means using mathematics to prove that code behaves exactly as intended, with no hidden flaws. It is the highest standard of security assurance, but it has traditionally been so slow and labour-intensive that most checks stop at the design stage, leaving a gap between what was designed and the code that actually runs on people’s phones. DeepMind says AlphaProof Code closes that gap.
The system works like a research team. Human experts first decide what must be proved and, working with a mix of AI models, translate Signal’s security goals into formal mathematical specifications, with experts reviewing every one. Then teams of AI agents powered by Gemini Carbon write the proofs themselves, producing more than six million lines of proof in Lean, a computer language used to check mathematical arguments. The agents then audited their own work and flagged just one percent of the code for a second look by humans, since the Lean system automatically checks the rest.
The properties proved include keeping messages protected even if a device is temporarily compromised, a safeguard known as forward secrecy and post-compromise security. The work covers the protocol’s one-to-one sessions, including its post-quantum handshake and its system for rotating message keys.
To test whether the system could tell good code from bad, DeepMind’s researchers reintroduced two real bugs that had previously been found and fixed in Signal’s code. In both cases, AlphaProof Code produced a machine-checked proof that the buggy version was insecure, showing it was not simply rubber-stamping the software.
Signal’s chief technology officer, Ehren Kret, said the work gives the company even greater confidence that its code delivers on its privacy guarantees. DeepMind has open-sourced the verification artifacts so outside researchers can inspect and build on them.
There are limits worth stating plainly. The proof covers the core cryptographic code for one-to-one chats, not the whole Signal app, and it does not cover group chats, other devices or every phone on which the app runs. Even so, DeepMind argues the method could eventually bring the same level of assurance to other critical software, from medical device firmware to power grid systems, and help defend against cyber threats that are themselves increasingly driven by AI.
Leave a Reply