The Daily Tech TipSmarter Tech. A Brighter You. ⌨ Home
security
security

How did AI mathematically prove Signal’s encryption code is correct — and what does it NOT prove?

By Adam Neely · Nov 11, 2026
Quick answer: Google DeepMind’s AlphaProof Code used Gemini Carbon-powered agents to formally verify key security properties of Signal’s compiled libsignal v0.105.4 code, producing over six million lines of Lean proof in four weeks and catching two reintroduced bugs. It does not prove the whole app unhackable: group messaging was out of scope and the proofs rest on stated assumptions.

Google DeepMind’s AlphaProof Code used Gemini Carbon-powered agents to formally verify key security properties of Signal’s compiled libsignal v0.105.4 code, producing over six million lines of Lean proof in four weeks and catching two reintroduced bugs. It does not prove the whole app unhackable: group messaging was out of scope and the proofs rest on stated assumptions.

Your group chat feels private, but “trust us, it is encrypted” is normally a promise, not a proof. Short answer: Google DeepMind’s AlphaProof Code, announced November 10, used AI agents powered by Gemini Carbon to formally verify key security properties of Signal’s actual compiled encryption library — libsignal v0.105.4 — generating over six million lines of Lean proof in four weeks. It checked the real shipped code, covering the PQXDH key agreement and the Triple Ratchet, across roughly 140,000 machine instructions — not a simplified model of it.Why it matters: normal software testing tries sample inputs and hopes the bad ones show up; formal verification is a mathematical proof that certain properties hold for every possible input, the way a geometry proof covers every triangle, not just the ones you drew.

As a sanity check, DeepMind reintroduced two previously fixed bugs into an isolated copy of the code, and the system caught both — evidence the proofs are actually watching the code, not rubber-stamping it. The proof artifacts are open-sourced, so outside researchers can re-check them.What it does NOT prove: it does not prove the whole Signal app is unhackable. Group messaging was out of scope, the proofs rest on stated assumptions about the hardware and surrounding code, and nothing here protects a phone that is already compromised.1.

Know the difference: testing samples behaviour; verification proves defined properties — ask which one a vendor actually did.2. Check your own messenger: look for published audits, open-source code, and whether the protocol — not just the app — has outside review.3. Keep the verified core reachable: install app and system updates promptly, since fixes only help once they are on your phone.Pro tip: verified core cryptography does not protect a compromised phone — a phone with an outdated OS or a malicious app already on it can leak messages before encryption ever gets a say.

Keep the OS updated and be picky about what you install.Ad’s take: this is the strongest kind of security claim there is — a proof strangers can re-run — and it still comes with honest limits printed on the box. That combination is exactly what I want from every app on my phone.

Open on the homepage

Ask AI about this tip

Send this tip to your favorite AI assistant for a plain-language walkthrough, follow-up questions, or help applying it to your device.

Was this helpful?
Home