Google DeepMind researchers say AI agents generated mathematical proofs connecting specified security guarantees to a compiled build of Signal's encryption library. The researchers published the evidence for outside verification. The November 10 AlphaProof Code report describes about 6.5 million lines of machine-checked proofs written in Lean. The authors say five people spent four weeks developing the proofs. Humans guided the process, chose reference models and reviewed the specifications defining the agents' proof tasks.
The verification targets an AArch64 Linux executable containing libsignal-protocol v0.105.4 and SPQR v1.6.0 within a small command-line harness. Libsignal implements the Signal protocol. The theorems address session initialization, message sending and message receiving. The verification excludes key generation.
The Google report links the executable to specified guarantees for PQXDH and Triple Ratchet one-to-one messaging. The guarantees include forward secrecy, authentication and post-compromise security. Those properties address the confidentiality of earlier messages after key theft, sender authenticity and the recovery of session secrecy after compromise. The results apply to the particular build under explicit cryptographic and operating-environment assumptions.
The symbolic model assumes idealized cryptographic operations and executions without collisions. The computational model places bounds on classical attackers' success under specified cryptographic assumptions. The symbolic proof establishes handshake secrecy under specified subsequent breaks of Diffie-Hellman. However, the computational proof excludes quantum attackers. The report excludes group messaging, and the computational model omits multi-device state. The computational bounds cease to offer useful guarantees when workloads become sufficiently large.
The results depend on trust in Lean's checking kernel, the researcher-reviewed specifications, the AArch64 Linux model and contracts for selected routines. A November 11 project update merged on GitHub says verification of ten additional library routines reduced trusted routine hooks from 37 to 27. Those hooks substitute assumed behavioral descriptions for execution of selected routines. The update keeps libsignal v0.105.4 but pins a different executable. The updated build verifies those ten routines at the machine-code level rather than depending on modeled contracts.
The repository includes public instructions for rebuilding the pinned binary and checking the proofs. A comparator available only on Linux checks the theorems and dependent definitions against trusted statements. The comparator then replays the proofs using a fresh Lean kernel. CryptoSlate has not independently rebuilt and checked the complete proof package.
The authors also say the researchers reintroduced two libsignal bugs that had previously been fixed. The authors report proving that the altered binaries breach the symbolic security theorem. That exercise tests whether the specification rejects those known regressions. The public package enables other researchers to examine the claimed link between executable code and security requirements.
Google AI Agents Audit Signal's Code, Major Flaw in Quantum Attack Coverage
CryptoSlateShare
Google AI agents audited Signal's code using FLASH AI, producing 6.5 million machine-checked proofs in Lean. The audit covered libsignal-protocol v0.105.4 and SPQR v1.6.0, verifying session management and message security under classical attacks. Forward secrecy and post-compromise security are confirmed, but quantum threats and group messaging are excluded. Researchers reinserted two bugs to test robustness. The package is public for verification, though CryptoSlate hasn't validated all proofs. The findings relate to CFT (Countering the Financing of Terrorism) and could affect liquidity and crypto markets.
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.



