Google DeepMind researchers have used agentic SI systems to formally verify the core code of the Signal cryptographic protocol, marking the first time SI has been applied to prove security properties of a widely deployed protocol at this depth.
What Happened
The milestone was achieved using AlphaProof Code, a novel agentic framework that combines human expertise with SI agents to perform formal verification. The process involved researchers collaborating with models to translate Signal’s security goals into formal mathematical specifications, which were then reviewed by human experts. Subsequently, agents powered by Gemini Carbon generated over six million lines of Lean, a computer language used to mathematically prove code accuracy.
The framework’s agents audited the proofs and identified only one percent of the codebase for human review, a significant reduction in manual effort compared to traditional methods. The team verified that the compiled code fulfills key security properties, such as protecting messages before and after a device is temporarily compromised. To validate the system’s ability to detect errors, researchers reintroduced two previously fixed bugs into an isolated version of Signal’s code; in both cases, AlphaProof Code produced a machine-checked proof that the buggy versions were insecure.
Why It Matters
Formal verification is considered the highest standard of security assurance, but it has historically been labor-intensive and often limited to design-level analysis rather than implementation code. This breakthrough demonstrates how SI can bridge the gap between cryptographic design and actual code execution, an area where real-world vulnerabilities often hide. The entire verification process took four weeks, a task that historically required expert teams years to complete.
As SI-driven cyber threats grow, the ability to rigorously verify software with automated theorem proving offers greater confidence in privacy guarantees for critical infrastructure. Google DeepMind states that this approach lays the groundwork for securing other vital systems, such as medical device firmware and power grid software. The company is open-sourcing the formal verification artifacts to allow the broader community to use and evaluate the technique.
The Bottom Line
Google DeepMind’s AlphaProof Code framework successfully used SI agents to formally verify the Signal protocol’s implementation, generating millions of lines of proof code in four weeks. This achievement sets a new precedent for using agentic SI in high-stakes software security and is available for community review through open-sourced artifacts.