Google DeepMind uses AI to verify Signal's encryption protocol
Researchers used a new AI framework to mathematically prove the security of the encryption used by Signal and several other major messaging apps.
The short version
- The four-week proof-of-concept project used agentic AI to scan for cryptographic flaws.
- No new vulnerabilities were found in the portions of the Signal protocol analyzed by the lab.
- Signal's encryption protocol is the foundation for security in apps like WhatsApp and Google Messages.
- DeepMind is releasing the specifications and proofs to help other researchers with formal verification.
- The analysis was a research project conducted at no cost to Signal.
Why it matters
This marks a milestone in using AI to mathematically prove that software is secure. It demonstrates how AI can speed up the process of verifying the privacy guarantees that protect global communications.
AI tests the gold standard
Google DeepMind recently completed a four-week proof-of-concept project to test the security of Signal, a messaging app widely considered one of the most secure in the world. The research team used agentic AI tools to formally verify several key cryptographic guarantees that make up Signal's underlying encryption protocol. Formal verification is a rigorous method used to mathematically prove that a piece of software behaves exactly as intended.
The researchers utilized a new AI framework called AlphaProof Code to scan the protocol for potential vulnerabilities. During the analysis, the AI did not discover any new flaws in the specific portions of the Signal encryption protocol that the lab examined. This result provides a high level of confidence in the privacy protections that the app offers to its users.
This project is significant because Signal's encryption protocol is not just for one app. It serves as the security foundation for several major platforms, including WhatsApp, Facebook Messenger, and Google Messages. By verifying the core protocol, the research validates the privacy infrastructure used by billions of people worldwide.
How AlphaProof Code works
Traditional formal verification is a notoriously slow and difficult process for human researchers. It requires humans to describe the characteristics of software they want to prove, translate those summaries into complex mathematical formulas, and then manually develop the proof. Because of this complexity, manual verification often focuses on only very small portions of a code base.
Google DeepMindās approach with AlphaProof Code aims to change this by using AI to handle the heavy lifting of the verification process. The AI agents can help significantly cut down the timeline of these projects by automating the transition from software descriptions to mathematical proofs. This allows for a significant expansion of what can be verified compared to previous human-led efforts.
The researchers noted that while the AI is powerful, it is not intended to replace humans entirely. They describe the future of this technology as a human-AI collaboration where AI agents contribute to the work and self-correct while humans provide oversight. This middle ground ensures that humans remain in control of the goals and the final review of the proofs.
A milestone for cryptography
The use of AlphaProof Code to mathematically prove parts of the Signal protocol is considered a major milestone in both the fields of cryptography and artificial intelligence. It demonstrates that AI can be used to rigorously verify the integrity of cryptographic protocols more efficiently than previous methods. This is especially relevant as AI-driven cyber attacks are increasing, threatening entire software supply chains and code bases.
When the DeepMind team tested older versions of the Signal protocol, the AI did identify security flaws. However, those specific flaws were already known to the public and had been patched by the developers long ago. The fact that the AI found known issues in old code but no new issues in the current version suggests the current protocol is robust.
Signal's chief technology officer, Ehren Kret, stated that this verification gives the team even greater confidence in their privacy guarantees. He noted that as AI-driven security threats grow, having a rigorous verification of the implementation is vital. The project serves as a proof that these tools can eventually be used to secure a wider variety of software.
Scope and future implications
While the results are promising, Google DeepMind clarified that the analysis was limited in scope. The study did not cover the entire Signal mobile app, and the findings were based on some past formal verification work already conducted by the Signal Foundation. There are also remaining questions about whether it is possible to automate formal verification at a massive scale at a reasonable cost.
Despite these limitations, the lab plans to release the specifications and proofs they developed for the Signal protocol analysis. By making these materials public, they hope to aid other researchers in their own formal verification work. This open approach helps build a shared foundation for verifying the security of other critical systems.
It is currently unclear if Google DeepMind plans to release a commercial product or service for formal verification in the future. For now, the project remains a research initiative conducted at no cost to Signal. It serves as a demonstration of how AI can be harnessed to protect the digital privacy of global communications.
Protecting global communications
The Signal protocol is a pioneer in the world of encrypted communications. It is favored by privacy activists because it collects minimal user data and offers features like hiding phone numbers. Unlike some other apps, Signal enables end-to-end encryption by default for all messages and voice calls, ensuring that no one can intercept the data in transit.
Since 2023, Signal has also been upgrading its technology to remain secure against the hypothetical threat of quantum computers. This ongoing work ensures that the protocol remains resilient as computing power evolves. The DeepMind verification adds another layer of assurance to these existing efforts by providing a mathematical proof of the current system's integrity.
The broader impact of this research extends to any company that relies on high-assurance security. Pushmeet Kohli, a vice president of research at Google Cloud and DeepMind, suggested that banks and other companies worried about bugs could eventually use these tools to prove their security properties. This could lead to a future where software security is verified by math rather than just tested for errors.
What this means for users
For the average user, this news means that the encryption protecting your private messages is being subjected to some of the most advanced testing available today. While you won't see a new button in your app, the underlying math that keeps your conversations private has been validated by a leading AI research lab. This provides peace of mind that the
What to watch
- Whether Google DeepMind releases a commercial tool for formal verification.
- How other major platforms like WhatsApp use these AI tools to verify their own protocols.
- The progress of AI-driven cyber attacks and how they might evolve to target these verified systems.
Try it
Check your messaging app settings to ensure end-to-end encryption is enabled by default.
Sources and links
First reported by The Indian Express. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.