DeepMind reports formal verification of Signal messaging protocol
The assessment covers the encryption protocol, not the entire Signal app. WhatsApp and Google Messages also use the protocol.
Google DeepMind researchers have used agentic AI to formally verify the security of the Signal protocol, which underpins encrypted messaging in Signal, WhatsApp and Google Messages. The analysis took four weeks and built on earlier formal verification work conducted without AI.
Formal verification uses mathematical methods to assess whether a protocol meets its stated security guarantees, rather than relying on tests that search for vulnerabilities. The researchers’ validation applies to the protocol itself, not to the Signal application in its entirety.
When applied to older Signal versions with known vulnerabilities, the approach detected those weaknesses and returned failed verification results.
Google told Wired that the computational demands leave uncertainty about how practical broader use would be. The amount of computing resources used for the four-week analysis has not been disclosed.
Swarat Chaudhuri, a DeepMind research scientist, told Wired that the result represents a milestone, but extending the approach to software more broadly remains a distant goal.
