Google DeepMind used the AlphaProof Code tool to verify the security of the Signal Protocol implementation down to the level of compiled code
New details: Google DeepMind's AI agents generated over 6 million lines of proofs in the Lean language, with humans reviewing only 1% of the work. The four-week project formally verified the security of the Signal Protocol at the level of compiled code, not just design, finding no new bugs.
Google DeepMind has clarified the scope of its work on the formal verification of the security of the Signal Protocol using the AlphaProof Code tool: according to the company, AI agents built on Gemini Carbon models generated over six million lines of code in the Lean language, which mathematically proves the correctness of the verified code. According to the company, human experts reviewed only about one percent of the resulting code – the rest was automatically checked by the Lean system. Verification took place at the level of code that was actually compiled and running on devices, not just at the level of the protocol's design, which, according to the company, is more difficult, because real vulnerabilities are often hidden precisely in the gap between design and implementation.
The formal verification project for the Signal Protocol itself, which was also reported on by outlets such as Wired and Ars Technica, took four weeks – a task that, according to Google DeepMind, has historically taken expert teams years. The team proved key security properties, for example the protection of messages before and after a temporary compromise of a device, and found no new vulnerabilities in the verified parts of the code. To test the correctness of the tool, researchers reintroduced two previously fixed real bugs into an isolated version of the code, and in both cases AlphaProof Code correctly produced a proof that such a version was insecure.
The Signal Foundation, through CTO Ehren Kret, stated that previous formal verification efforts focused on smaller parts of the code and that Google DeepMind's approach allows for a significant expansion of scope, which gives the company greater confidence regarding the privacy protection guarantees in the code. According to Ars Technica and Wired, researchers emphasize that this is a proof-of-concept, not a verification of the entire Signal mobile application, and that it remains an open question whether this approach can be scaled to large amounts of software and at what cost – Google has not disclosed the project's costs and has not commented on any potential commercial plans.
Why it matters
It turns out that agentic AI can shorten the formal verification of security-critical code from years to weeks while also reducing the need for human review to a fraction of the codebase – this could change how the security of cryptographic implementations is examined for services with massive usage, such as Signal, Google Messages, or WhatsApp. At the same time, this is still a one-off research project with no disclosed price and no commercial tool, so practical availability and scalability for ordinary companies has not yet been demonstrated.
What was added since the original report
Verified updates
-
Two deliberately planted bugs in the code were caught immediately during verification; DeepMind plans to extend the AlphaProof Code method to power grids and medical devices; Signal's CTO confirmed that the approach strengthens confidence in data privacy protection
- Two deliberately planted bugs in the code were caught immediately during verification
- DeepMind plans to extend the AlphaProof Code method to power grids and medical devices
- Signal's CTO confirmed that the approach strengthens confidence in data privacy protection
-
AI agents generated over 6 million lines of Lean code as part of the verification; Humans manually reviewed only 1% of the codebase; Verification took place at the level of the compiled code implementation, not just at the design level; The framework combines human expertise with the autonomous work of AI agents; This is the first time a cryptographic protocol has been formally verified at the implementation level by AI
- AI agents generated over 6 million lines of Lean code as part of the verification
- Humans manually reviewed only 1% of the codebase
- Verification took place at the level of the compiled code implementation, not just at the design level
- The framework combines human expertise with the autonomous work of AI agents
- This is the first time a cryptographic protocol has been formally verified at the implementation level by AI
Two audiences, two different impacts
What this means
For individuals
For developers and security researchers working with formal methods, it is relevant that the publication of the specification and proof for the Signal Protocol will provide concrete reference material for further verification and study of agentic verification procedures.
For a business
Companies that operate or audit cryptographic implementations (messaging, banking) may want to watch whether Google DeepMind expands AlphaProof Code into a commercial formal-verification offering – for now it is a free research project with no disclosed costs and no confirmed productization plan.
Risks and complianceCheck the original
Event sources
confirmed by 2 independent sources · 3 publishers, 2 independent. We count feeds from the same owner only once.