Google DeepMind announced AlphaProof Code, a system of AI agents that can help formally verify software. Formal verification means using mathematics to prove that code does what it is supposed to do, with very high certainty. The team used it on the actual code of the Signal messaging protocol (the encryption system used by Signal and, in related forms, by apps like WhatsApp and Google Messages).
In about four weeks, AI agents (guided by human experts) produced roughly 6.5 million lines of proof code that was checked by a reliable mathematical proof system called Lean. The proofs cover important security properties such as forward secrecy (old messages stay safe even if keys are later compromised), authentication, and post-compromise security. Importantly, the verification was done on the real compiled machine code of the Signal library, not just on a simplified model or rewritten version.
This is significant because full formal verification of complex real-world code has traditionally taken many years of expert work. AI agents helped speed up the proof-writing part while humans defined what needed to be proved and reviewed the specifications. The proofs and related materials are being open-sourced. For students studying computer science, cybersecurity, or AI, this is a clear example of how agentic AI can assist with deep technical security work that was previously extremely difficult.
Source: Deepmind Google.