Safety · 2026-11-10

Google DeepMind used AI to mathematically verify the security of the Signal protocol

In four weeks, Google DeepMind researchers used AI agents to prove that key cryptographic guarantees of the protocol protecting Signal messages work as intended.

✨ What the researchers did

The Google DeepMind team used AI agents to mathematically verify key cryptographic guarantees of the Signal protocol. This approach is called formal verification. It means checking that software works correctly using math, not just tests.

The project was a proof of concept and lasted only four weeks. For the verification, the team used a new AI tool called AlphaProof Code.

📌 Why this matters for ordinary people

The Signal protocol protects a huge number of messages. Its cryptography is also used in Google Messages and WhatsApp. Independent verification of this cryptography adds confidence that privacy works as intended.

The researchers did not find new vulnerabilities in the parts of the protocol they examined. However, when they ran the tool on older versions of Signal with known and long-fixed vulnerabilities, it correctly identified them and showed that those versions could not be formally verified.

🔹 What is still unknown

The analysis did not cover the entire Signal app, only certain parts. The researchers stress that this is just one step. It is still unclear whether such verification can scale to all software and whether it will be affordable. Google did not disclose the cost of the analysis and did not say whether it plans a commercial formal verification product.

The team also stresses that combining humans and AI was key. The agents propose proofs and correct themselves, but humans remain involved at important stages. As researcher Swarat Chaudhuri put it, there should be a middle ground, not a world where AI sets the goals, writes the proofs, and grades its own tests.

Sources: 🔗 WIRED — AI🔗 Google DeepMind Blog🔗 DOU/dev.ua🔗 Ars Technica — AI