Google AI agents audit Signals code and one major flaw stays hidden

Extracto:Google DeepMind's AI agents generated 6.5 million lines of machine-checked proofs verifying security properties in Signal's code.

Google DeepMind researchers say their AI agents produced mathematical proofs linking specific security guarantees to a compiled build of Signals encryption library, and they put the resulting evidence online for others to check.

Related Company Google An American multinational technology company

The Nov. 10 AlphaProof Code report describes about 6.5 million lines of machine-checked Lean proof code. The result concerns a particular build of libsignal, the Signal protocols implementation, under explicit assumptions about cryptography and the environment in which the code runs.

The authors say five people worked on the proofs over four weeks. Humans directed the work, selected reference models and reviewed the specifications that defined what the agents had to prove.

What Google DeepMind's proofs cover

The target was an AArch64 Linux executable using libsignal-protocol v0.105.4 and SPQR v1.6.0 inside a small command-line harness. The theorems cover session initialization, sending a message, and receiving one. Key generation falls outside that verified scope.

The Google report connects the compiled executable to specified properties of PQXDH and Triple Ratchet one-to-one messaging, including forward secrecy, authentication, and post-compromise security.

These address whether earlier messages stay secret after keys are stolen, whether senders are authentic, and whether a session can regain secrecy after a compromise.

Its two proof models provide different assurances: the symbolic analysis relies on idealized cryptographic operations and collision-free executions, while the computational analysis bounds the success of classical attackers under stated cryptographic assumptions.

What a checked Signal proof establishes

DeepMinds reported process connects human-selected requirements and one executable to a machine-checked proof, with trusted assumptions still defining the result.

Specification and executable feed an AI proof, which Lean checks under trusted assumptionsTWO INPUTSHuman-reviewedspecificationDefines what must be trueOne compiled buildAArch64 Linux libsignal-cliInitialize · send · receive

AI writesthe proofAgainst the specification

Leanand comparatorCHECKING GATEProof matches thestated theorem

Specified guaranteeFor this build and modelTrusted basisSpecification · machine model · routine contracts · cryptography

  • SCOPE BOUNDARYNo direct proof of the phone apps, key generation or group messaging.

How the reported Signal proof depends on its specification and buildTWO INPUTSHuman-reviewedspecificationDefines whatmust be trueOne compiledbuildAArch64 Linuxlibsignal-cliInitialize · send· receive

AI writes the proofAgainst the specification

Leanand comparatorProof matches the stated theorem

Specified guaranteeFor this build and model

Trusted basisSpecification · machine modelRoutine contracts · cryptographySCOPE BOUNDARYNo direct proof of phone apps,key generation or group messaging.

A checked proof establishes the stated guarantee for the specified build and model.

Author-reported proofs for a specific AArch64 Linux harness. Specifications, execution model, routine contracts and cryptographic assumptions remain trusted. No direct verification of phone apps, key generation or groups; no independent rebuild performed here.

Sources and evidence

Sources

  • AlphaProof Code technical report
  • Public proof and comparator instructions
  • Supporting facts

    • Reported statement In a technical report dated November 10, 2026, Google DeepMind researchers describe AlphaProof Code, a human-guided multi-agent framework that produced about 6.5 million lines of machine-checked Lean proof code for a compiled libsignal command-line build. [1]
    • Reported statement The report says five people worked on the proofs over four weeks, directing agents, selecting reference models and reviewing specifications; the paper lists nine authors. [1]
    • Reported statement The reports verification target is an AArch64 Linux executable built from libsignal-protocol v0.105.4 and SPQR v1.6.0 in a libsignal-cli harness; its theorems cover session initialization, sending and receiving, but not key generation. [1] [2]
    • Reported statement The report‘s symbolic proof establishes handshake secrecy under specified later Diffie-Hellman breaks, while the computational proof excludes quantum attackers and multi-device state; group messaging is outside the report’s scope. [1]
    • Reported statement The authors‘ guarantees concern their libsignal-cli build and modeled execution environment, not direct verification of Signal’s complete phone applications. They depend on specifications reviewed by humans, Leans checking kernel, the AArch64 Linux model, contracts for selected routines and assumptions about the underlying cryptographic primitives. [1]
    • Reported statement The project repository provides instructions to rebuild a pinned binary, check Lean proofs and run a Linux-only comparator that checks theorem statements and their dependent definitions against trusted statements and replays proofs in a fresh kernel. [2]

    Method

    Source-bound schematic of the reported proof workflow and trust boundary; connectors show logical dependence, not elapsed time or measured security strength.

    Scope

    Author-reported proofs for a specific AArch64 Linux harness. Specifications, execution model, routine contracts and cryptographic assumptions remain trusted. No direct verification of phone apps, key generation or groups; no independent rebuild performed here.

    Infographic by CryptoSlate

    The symbolic proof establishes handshake secrecy under specified later breaks of Diffie-Hellman, but the computational proof does not cover quantum attackers. Group messaging is outside the reports scope, and the computational model excludes multi-device state. Its bounds also stop providing a useful guarantee at sufficiently large workloads.

    The results still require trust in Leans checking kernel, the specifications reviewed by the researchers, the AArch64 Linux model, and contracts describing selected routines.

    The projects Nov. 11 update, merged on GitHub, says ten more library routines were verified, cutting trusted routine hooks from 37 to 27. Those hooks replace execution of selected routines with assumed descriptions of their behavior.

    The update retains libsignal v0.105.4 but changes the pinned executable. The newer build verifies the machine code for those ten routines instead of relying on their modeled contracts.

    The project repository provides public instructions to rebuild the pinned binary and check its proofs. A Linux-only comparator checks that the theorems and their dependent definitions match trusted statements, then replays proofs in a fresh Lean kernel.

    The authors also report reintroducing two previously fixed libsignal bugs and proving that the modified binaries violate their symbolic security theorem. That tests whether the specification rejects those known regressions.

    The public package gives other researchers a concrete way to check the claimed connection between security requirements and executable code.

Descargo de responsabilidad

Las opiniones de este artículo solo representan las opiniones personales del autor y no constituyen un consejo de inversión para esta plataforma. Esta plataforma no garantiza la precisión, integridad y actualidad de la información del artículo, ni es responsable de ninguna pérdida causada por el uso o la confianza en la información del artículo.
El anterior

Más del 35% del suministro de Bitcoin enfrenta riesgo cuántico, según nuevos datos

El siguiente

¿Sufrirá Bitcoin una volatilidad drástica la noche de los resultados de Nvidia? El mercado de predicciones de WikiBit lanza el "3% Challenge"