DeepMind AlphaProof Code verifies Signal protocol in four weeks
Google DeepMind introduced AlphaProof Code, an agentic framework that formally verified the security properties of the Signal protocol's actual code in four weeks.
Google DeepMind researchers introduced AlphaProof Code, a novel agentic framework for formal verification. The system was used to mathematically prove key security properties of the Signal protocol's actual code. This work marks the first time AI has been used to formally verify a widely deployed cryptographic protocol at this depth. Historically, expert teams take years to perform the same verification task, and earlier efforts stopped at the theoretical design stage. In contrast, the verification process for Signal's core cryptographic code took just four weeks.
AlphaProof Code is a multi-LLM agentic framework that splits verification work between human experts and AI agents. Researchers collaborated with models to translate Signal's security goals into formal mathematical specifications reviewed by human experts. AI agents powered by Gemini Carbon generated over six million lines of Lean code for the proofs. The AI agents audited the proof and highlighted areas needing human review for just one percent of the codebase. The system produced machine-checked proofs showing that two reintroduced buggy versions of Signal's code were insecure.
Signal CTO Ehren Kret stated that the approach strengthens trust in data privacy as AI threats grow. Google DeepMind plans to extend the verification method to power grids and medical devices.
Sources
- Google DeepMind BlogIntroducing AlphaProof Code: a breakthrough in verifying complex software at depth ↗
- The DecoderGoogle Deepmind's AlphaProof Code verified Signal's cryptographic core in four weeks instead of years ↗
Every sentence of this story was checked against facts quoted from these articles. How we make the news
Spotted an error? Tell us
