DeepMind's AlphaProof Code formally verifies Signal's libsignal

Google DeepMind announced on November 10, 2026, that its multi-agent formal verification system AlphaProof Code has formally verified libsignal, Signal's core encryption library. According to the company's official blog, the system produced more than 6.5 million lines of mathematical proof code over four weeks.

The verification covered Signal's PQXDH handshake and its Triple Ratchet mechanism used for end-to-end encryption, showing that confidentiality, authentication, forward secrecy and post-compromise security hold in the compiled machine code. Security properties that assume quantum computer attacks were verified separately in a mathematical model.

As a test, researchers reinserted two previously fixed Signal bugs into the code. AlphaProof Code proved mathematically that neither satisfied the security conditions. The work also narrowed the code areas requiring additional human review to about 1% of the total.

Google DeepMind released the mathematical specifications and proofs used in the verification as open source on GitHub. Signal CTO Aaron Kretz said AI agents greatly expanded formal verification from a limited scope to actual implementation code.

Scope and limits

The verification was limited to one-to-one messaging encryption code compiled in a specific execution environment. It did not cover group messaging or multi-device state.

The result follows other cryptography-adjacent research this month. On November 6, Google said AI had found two weaknesses in RSA encryption and published an algorithm cutting the cost of factoring encryption keys by up to 30 times; earlier in November, Anthropic released FlashProve, which cryptographically proves that an AI model's response came from a specific model. Signal's protocol or encryption based on it is used by messengers including WhatsApp and Google Messages.