Google AI agents audit Signal’s code and one major flaw stays hidden

Google AI agents audit Signal’s code and one major flaw stays hidden

Summary

Google DeepMind researchers used AI agents, guided and reviewed by humans, to create about 6.5 million lines of machine-checked Lean proofs connecting specified security guarantees to a pinned compiled build of Signal’s libsignal. The proofs cover one-to-one session setup and messaging, including properties related to forward secrecy, authentication and post-compromise security, under stated assumptions. They do not cover areas such as group messaging or quantum attackers in the computational model. The work depends on trusted specifications, models and selected routine contracts; a later update verified ten additional routines, reducing those assumptions. Public tools let researchers rebuild the executable and check the proofs. The team also showed that binaries reintroducing two known bugs violate the symbolic security theorem.