Google used AI to verify Signal's encryption: 6 million lines of mathematical proof in four weeks
Direct answer
On November 10, 2026, Google DeepMind announced that its AlphaProof Code framework verified key security properties of the real code behind Signal's encryption protocol. The work produced more than 6 million lines of mathematical proof in four weeks, checking specified properties of the code. The researchers also confirmed the system rejected older versions with known bugs. The verification covers specified properties and code, with the declared scope.
What was verified and what was not
- The real code running on phones: AlphaProof Code worked on the code Signal actually uses to encrypt private messages. The mathematical proof speaks about the software running on phones.
- Specified properties: the verification covers concrete security properties the researchers defined in advance. It is a proof with a declared scope: what was verified stays verified; what stayed outside the scope stays unverified.
- Four weeks of work: the full analysis took four weeks. That is a short timeline for a formal verification of this size, and it shows AI can do in weeks work that used to take months of human experts.
- The known-bug test: the researchers confirmed the system rejected older versions with known bugs. A tool that accepts everything is useless; one that catches what is already known to be wrong shows it is really looking.
Why this matters beyond Signal
- AI checks AI: AI models write more and more code, and formal verification is the way to check that code with mathematics, with proofs anyone can review. AlphaProof Code is an example of AI doing the review work humans cannot do at scale.
- Security becomes demonstrable: a company that says its product is secure makes a promise. A company that publishes a formal verification with 6 million lines of proof shows its work. Enterprise customers are starting to ask for the second.
- Messaging encryption is critical infrastructure: Signal protects the private messages of millions of people. Verifying its protocol with formal proofs raises the standard for the whole sector: other messaging apps now have an example of what can be demonstrated.
- Formal verification reaches businesses: the same technique applies to code that handles payments, customer data and agent decisions. The company that verifies its critical code has an advantage when a customer or a regulator asks how it knows it works.
What to do in your business this week
- Ask your vendors how they verify their code: when an AI vendor talks about security, ask for the evidence: audits, formal proofs, independent reports. A promise without evidence is a marketing line.
- Identify your critical code: code that touches money, customer data or agent decisions deserves formal review. The rest can keep the usual tests. Formal verification is expensive; use it where failure costs the most.
- Keep the proofs alongside the results: a formal verification produces a record that can be reviewed. That record is the document you show when a customer, an auditor or a regulator asks.
- Follow the standard that is forming: DeepMind published this work as a research framework. The security teams studying it today are the ones applying it tomorrow.
Note: this article summarizes Google DeepMind's announcement of November 10, 2026 and coverage in The Neuron and industry roundups. The verification covers specified properties and code of Signal's encryption protocol, with more than 6 million lines of mathematical proof in four weeks. It does not cover every part of the system or every possible attack.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- AI found two flaws in RSA-1024 encryption
- OpenAI launched GPT-6 Cyber: it finds 81 new Firefox flaws and reproduces 948 Linux exploits
- Privacy and data: what happens to your customers' information
Frequently asked questions
What did DeepMind verify in Signal?
Google DeepMind's AlphaProof Code framework verified key security properties of the real code behind Signal's encryption protocol, with more than 6 million lines of mathematical proof in four weeks, according to the November 10, 2026 announcement.
Does the verification cover all of Signal?
It covers specified properties and code. The declared scope is part of the verification's value.
Why did it take only four weeks?
AI generated the mathematical proofs at a scale human work cannot reach in that time. That is the central point of the announcement: formal verification, which used to take months of experts, speeds up with AI.
What does this mean for my business?
Provable security through formal proofs is becoming the standard. When you evaluate AI vendors, ask for verification evidence, audits and independent reports.
