Artificial Intelligence

'Signal Shot' Reaches 80% in Signal Verification Project

Is AI making secure software more achievable or is SignalShot still far from its goal?

13 hours ago·Updated 1 hour ago

'Signal Shot' Reaches 80% in Signal Verification Project

Story

Above: Signal Messenger's logo displayed on a smartphone screen in this photo illustration, June 15, 2025. Image credit: Thomas Fuller/SOPA Images/LightRocket/Getty Images

The Facts

SignalShot, which launched in April as a community quest to formally verify Signal Messenger, is roughly 80% complete, according to a project status update Wednesday. The Beneficial AI Foundation (BAIF) and Google DeepMind have made roughly equal and complementary contributions.
Max Tegmark on X
Beneficial AI Foundation
Google DeepMind said its team formally verified a real-world implementation of the Signal cryptographic protocol at the machine-code level using AlphaProof Code, an AI framework for formal specification and verification released with a paper and Lean code.
Swarat Chaudhuri on X
Google DeepMind
Beneficial AI Foundation
Over four weeks in October, AlphaProof Code agents produced more than 6 million lines of machine-checked Lean proof. The Google DeepMind team stated that it proved "key security properties of libsignal" down to the ARM machine-code level.
Swarat Chaudhuri on X
Beneficial AI Foundation
SignalShot launched as an open-source collaboration led by the Beneficial AI Foundation and the team behind the Lean theorem prover, betting that rapid AI advances would automate much of the work of making Signal provably secure.
Beneficial AI Foundation
Lean on Youtube
Wednesday's SignalShot update cautioned against "formal-washing" — overhyping what has been proven — and said headlines suggesting Signal is now fully secure, including Wired coverage, could reduce researchers' motivation to finish the remaining work.
Max Tegmark on X
Beneficial AI Foundation
Wired
Outstanding tasks include verifying Signal's anonymous credential system, independently validating about 26,000 lines of trusted Lean code and building continuous verification. DeepMind said it released its specs and proofs for community improvement.
Beneficial AI Foundation
Swarat Chaudhuri on X

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Narrative A

SignalShot's 80% progress marks a major step toward making mathematical security verification practical at scale. Google DeepMind's AI agents generated more than 6 million lines of machine-checked proof in just four weeks, accomplishing work that historically could take expert teams years. By combining automated reasoning with human expertise and open-source collaboration, the project shows how AI can make provably secure software increasingly achievable.

Swarat Chaudhuri on X

Google DeepMind

Narrative B

SignalShot's 80% progress is encouraging, but the remaining 20% is essential to its goal of making Signal provably secure. Anonymous credentials, unfinished messaging protocols, 26,000 lines of trusted Lean code, other chip architectures and the latest Rust implementations still need verification. Even completed proofs must be independently validated, and continuous verification is needed to ensure future updates do not undo the work. The job is far from finished.