Most of the AI news cycle is about chatbots, but this one is quietly a big deal. Google DeepMind researchers say they've formally verified the core security of Signal, the encrypted messaging app used by journalists, activists, and many everyday people. In plain terms, they built a mathematical proof that Signal's code behaves the way its designers promised.
The tool is called AlphaProof Code. It's a team of AI agents, guided by humans, that writes a precise description of what the software should do, checks it against the real code, and then produces a proof that a separate system called Lean can verify. The team focused on two parts of Signal: its post-quantum handshake (PQXDH), meant to resist future quantum computers, and the Triple Ratchet, which keeps changing encryption keys so a single leak doesn't expose a whole conversation.
The numbers are wild. The agents generated around 6.5 million lines of Lean code. The whole effort took five people about four weeks. For comparison, The Register notes that famous earlier verification projects like the seL4 operating system kernel took about 20 person-years. Swarat Chaudhuri, a DeepMind research scientist and UT Austin professor, said the system is built for humans and AI working together, and that the next frontier is writing good specifications, not the proofs themselves.
There's a catch worth knowing. The proof assumes the underlying cryptography is perfect, so it isn't a guarantee that nothing could ever go wrong. Still, if AI can shrink years of verification work to weeks, more of the software we depend on could get checked properly.
Would you trust an app more if an AI had mathematically proven it secure, or does that make you more nervous?
