# Google DeepMind's AI proves key Signal protocol guarantees in four weeks > Source: > Published: 2026-11-10 08:05:51+00:00 *If your product leans on Signal-style encryption, a machine-checked proof is a stronger claim than an audit report. Read the scope before you repeat it in a pitch deck.* Google DeepMind used an AI system called AlphaProof Code to formally verify selected security guarantees of the Signal protocol, the encryption design behind Signal and, in modified form, Google Messages and WhatsApp. WIRED reported the result this week, and the researchers say it took four weeks and turned up no new vulnerabilities in the parts they examined. It was a proof of concept, not a clean bill of health for a messaging app. According to the coverage of the WIRED piece, DeepMind did not verify the whole Signal mobile app, and the work checked chosen protocol guarantees rather than every part of those apps or every way they might fail. A formal proof is a different animal from a code review. A reviewer reads the design and says it looks sound. A proof is checked by software, line by line, against a written specification, so the claim is only as good as that specification and its assumptions. The reports we found don't spell out which components were covered or which proof system the checker used, and DeepMind's own write-up wasn't something we could confirm. Treat any claim of "Signal is proven secure" as shorthand for "these stated properties hold under these stated assumptions." One detail makes the result more credible than a victory lap would be. Reports on the WIRED story say the team also pointed the tool at older Signal versions with known, long-fixed vulnerabilities. It found them, and it showed those versions could not be formally verified. A verifier that fails the broken code is doing its job. [Signal's Meredith Whittaker says AI agents are surveillance infrastructure and she's right](https://startupfortune.com/signals-meredith-whittaker-says-ai-agents-are-surveillance-infrastructure-and-shes-right/) Signal president Meredith Whittaker issued a blunt warning on June 20 that AI agents like Microsoft Copilot are surveillance infrastructure in disguise, arguing their sweeping access to contacts, messages, and credentials effectively nullifies end-to-end encryption. Her critique lands at a pivotal moment for startups building agentic workflows on... - [AI agents privacy surveillance concerns](https://startupfortune.com/signals-meredith-whittaker-says-ai-agents-are-surveillance-infrastructure-and-shes-right/) - [how to protect data from AI](https://startupfortune.com/signals-meredith-whittaker-says-ai-agents-are-surveillance-infrastructure-and-shes-right/) The humans stayed in the loop. DeepMind research scientist Swarat Chaudhuri told WIRED, "We are very excited by this result, but there is a long distance to go before we are able to scale this to all the software of the world, and we want to do that." He also said, "We don't want this sort of vibe world where AI is generating the goals and AI is generating the proofs," according to Superpower Daily's account. Signal Foundation technology director Ehren Kret said confirming the protocol's privacy guarantees matters more in an era of AI-driven threats, per Kantan News. DeepMind has released the specification and the proof so others can build on them. Google declined to disclose what the project cost or say whether it plans to sell the tool, which leaves the economics unanswered. ## Why the old guarantees still needed this Nobody expected the design to collapse. Academic analyses, including Oxford's 2017 work on the double ratchet, found no major flaws in the protocol. What has been scarce is proof that the code people actually run matches the design on paper. That is the gap the Beneficial AI Foundation, co-founded by Max Tegmark, is attacking with Signal Shot, a Lean-based effort with Signal and the Lean FRO to formally verify the protocol and its Rust implementation. DeepMind's result lands in the same territory from a different direction, and it is independent of that project. Frankly, this is the more interesting story than the headline. Four weeks is a short time for a task that has historically taken specialist teams much longer, and a proof an AI system helped write is only worth something because a separate checker can reject it. ## What it means if you ship security software If you run a crypto wallet, a messaging startup or anything that sells encryption, the direct lesson isn't that Signal is safe. People already believed that. It's that machine-checked proofs are getting cheap enough to ask for, and your customers and auditors will start asking. A vendor that can point to a published specification and a verified proof has a real edge over one that says "audited by a reputable firm." The limits are just as practical. Signal's protocol is small, heavily studied and written down in public documents, which makes it close to the best-case target. Most commercial code is bigger, messier and has no clean spec to prove against. Chaudhuri himself says scaling to all software is a long way off, and whether the method pays for itself on ordinary commercial systems is still an open question. The specification and the proof are public, so anyone with the right tools can now check DeepMind's work rather than take it on trust. For a result about trust, that's the right way round. [Google DeepMind's AI Control Roadmap treats its own agents as insider threats and sets the compliance bar for everyone else](https://startupfortune.com/google-deepminds-ai-control-roadmap-treats-its-own-agents-as-insider-threats-and-sets-the-compliance-bar-for-everyone-else/) Google DeepMind published a formal AI Control Roadmap on June 18, treating its own advanced agents, including Gemini Spark, as potential insider threats requiring containment layers beyond alignment training. Built on analysis of over one million coding tasks and adapted from the MITRE ATT&CK framework, the 15-control stack sets a compliance... - [how to control AI agents internally](https://startupfortune.com/google-deepminds-ai-control-roadmap-treats-its-own-agents-as-insider-threats-and-sets-the-compliance-bar-for-everyone-else/) - [preventing AI insider threat risks](https://startupfortune.com/google-deepminds-ai-control-roadmap-treats-its-own-agents-as-insider-threats-and-sets-the-compliance-bar-for-everyone-else/) **Also read:** [Rapidus is likely to miss its 2nm mass-production deadline, Bloomberg reports](https://startupfortune.com/rapidus-is-likely-to-miss-its-2nm-mass-production-deadline-bloomberg-reports/) • [Australia plans an AI safety regulator and mandatory incident reporting for frontier labs](https://startupfortune.com/australia-plans-an-ai-safety-regulator-and-mandatory-incident-reporting-for-frontier-labs/) • [Apple starts selling $50 Intelligence Usage Packs for AI features in Creator Studio](https://startupfortune.com/apple-starts-selling-50-intelligence-usage-packs-for-ai-features-in-creator-studio/) *This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.* ## Join the discussion [Open in the community →](https://startupfortune.com/community/) Almost there. Sign in and your reply posts straight away.