Google DeepMind and Google researchers introduced AlphaProof Code on 10 November 2026, reporting machine-code proofs of selected security properties of libsignal, the Signal protocol’s reference implementation. The framework combines human-reviewed Lean specifications with autonomous proof agents; Lean’s kernel checks the resulting mathematical proofs.[1]
The authors say they linked unmodified libsignal into a small command-line harness and compiled it with stock rustc for AArch64 Linux. They report about 6.5 million lines of Lean proofs, resting on roughly 26,000 lines of trusted specifications and machine-code semantics. A comparator checks that theorem statements and their dependent definitions match an approved, held-out specification, addressing the risk that proof generation silently weakens the claim.[1]
The reported results cover PQXDH and Triple Ratchet. The introduction describes confidentiality, authenticity, forward secrecy and post-compromise security against classical attackers, plus symbolic handshake secrecy against a future quantum attacker. These claims have explicit boundaries: cryptographic primitives are idealized or assumed secure, and some called routines use trusted contracts. The execution model covers a single-threaded, statically linked process and excludes features including threads and dynamic linking.[1]
[1]: Google DeepMind and Google, “AlphaProof Code: Verifying Signal’s Security at the Machine-Code Level,” 10 November 2026. https://google-deepmind.github.io/code-verification/
Sources
Check or correct this page
Verification record
The digest below shows whether the text you are reading matches the published source. It does not show that the reporting is correct. The build computes it from this page's Markdown source and lists it in the site manifest.
371041b8e175565269db718cdef8bbb51e0e41dda555453894f536828979d074