Momentum in Provable Security: From libsignal to High-Assurance Infrastructure

The Beneficial AI Foundation recently shared an update on the Google Deep Mind team’s progress proving key security properties in libsignal. It’s great to see the community continuing to tackle hard, high-stakes challenges in provable security – a field Galois has been proud to help shape, including recent work with Signal.

“We formally verified the security of Signal's PQXDH,” said Galois Principal Scientist Tom Shrimpton. “Using Aeneas, we extracted a Lean model of libSignal's Rust implementation. We then defined authenticated key-exchange security definitions in VCVio (which have now been adopted and integrated into VCVio) and then leveraged AI to develop the formal proof that the code-extracted model of PQXDH satisfied the security definition.”

Alongside this verification work, Galois also created a resource to bridge the gap between the Signal’s public protocol documentation and the libsignal codebase that implements it. 

Beyond Signal, Galois has been driving impactful work in provable security for decades –  using applied mathematics to guarantee that particular classes of security flaws cannot happen.

Since 2015, Galois has worked with AWS to help ensure the security of some of the key cryptographic systems that protect their cloud. And in 2020, we partnered with Supranational to provide formal verification for ‘blst’, a BLS signature library that serves as an important part of consensus protocols for next-gen blockchain technologies like Eth2 and Filecoin.

Together, these collaborations yielded the highest possible level of continuous protection for a large percentage of the world’s most valuable and most critical data. That means that Galois has played a key role in securing financial transactions that keep the global economy churning, state secrets that keep citizens safe, and the privacy of your and my personal communications. This is no small thing, it’s real, tangible impact at the highest of levels – all thanks to applied mathematics.

More recently, in 2026 we announced that SAW now supports generating Isabelle theories from Cryptol specifications – a new capability developed by Galois in collaboration with Apple. This ability to translate from Cryptol to Isabelle within SAW represents a remarkable leap forward – significantly expanding the scope of the possible in software verification. The classic usability–efficacy tradeoff in formal methods has long been taken as a given. That assumption is starting to crack. 

Our work in this space isn’t about chasing shiny objects. Every project is another brick in the foundation of a larger vision: a future built on secure, high-assurance infrastructure.