DeepMind’s Signal Proof Is a Security Breakthrough With a Big Scaling Question
DeepMind’s Signal Proof Is a Security Breakthrough With a Big Scaling Question
In a four-week research effort, Google DeepMind paired human experts with Gemini-powered agents to turn Signal’s security goals into mathematical specifications, then prove key properties of the compiled code that runs on devices. The company says AlphaProof Code generated more than six million lines of Lean proof code, while humans reviewed the goals and the small share of work flagged for scrutiny.
The distinction matters. Formal checks of cryptographic designs are not new, but DeepMind says its system bridged the perilous gap between a protocol’s intended design and its real implementation — where vulnerabilities can lurk. To test whether the system could distinguish secure code from flawed code, researchers reintroduced two previously fixed Signal bugs. In both cases, the framework produced machine-checked evidence that the altered version was insecure.
After the results were announced, DeepMind chief executive Demis Hassabis amplified the claim that the work marked the first formal verification of the actual production code of a widely deployed protocol. Google chief executive Sundar Pichai stressed that the agents verified the protocol “down to compiled device code,” portraying the project as a way to close the gap between security designs and the software users actually run.
Signal’s chief technology officer, Ehren Kret, welcomed the result, saying the approach enables a “significant expansion” beyond earlier verification work on smaller slices of code and gives the organization greater confidence in its privacy guarantees.
Yet the celebration comes with a sharp qualification. The work covered key parts of Signal’s protocol, not the entire mobile app, and it was conducted as a free research project rather than a commercial service. DeepMind researcher Swarat Chaudhuri told WIRED there is “a long distance to go” before the technique can scale to all software. The proof is striking; turning it into routine protection for the world’s code remains the unfinished challenge.
Write a comment