Google DeepMind researchers have shown that an AI system can mathematically prove important security properties of the Signal messaging protocol. The team used a new framework called AlphaProof Code to formally verify several cryptographic guarantees. The work took about four weeks and was done as a research project, not a commercial product. No new bugs were found in the sections they examined, and the AI correctly flagged older, already‑fixed flaws when it was run on previous versions of the protocol.
Formal verification means using math to check that software behaves exactly as intended, leaving no room for hidden vulnerabilities. Historically this process required experts to write detailed specifications, translate them into formulas, and then construct the proof—a labor‑intensive task that could take months. By letting an agentic AI (an AI that can set and pursue its own sub‑goals) handle much of the heavy lifting, DeepMind cut the time down to weeks. The researchers say the AI still needed human oversight at key steps, but it could suggest and self‑correct proofs on its own.
Signal’s chief technology officer, Ehren Kret, noted that prior verification work covered only small code fragments. DeepMind’s approach, he said, allows “significant expansion” of what can be checked. The team is releasing the specification and proof they built so other groups can build on it. While the result is promising, DeepMind cautions that scaling this to entire apps or whole software supply chains remains a challenge, and the cost of such AI‑driven verification has not been disclosed.
Why this matters
If AI can reliably prove the security of cryptographic protocols, everyday users of Signal, Google Messages, WhatsApp and similar apps could gain stronger confidence that their messages stay private. For businesses that depend on encrypted communication—banks, healthcare providers, and journalists—the technology could eventually provide legally verifiable guarantees that their software has no critical flaws.
Why it matters
For the average person, the news means that the encryption protecting their chats is now backed by an additional mathematical check, reducing the chance of hidden bugs. For companies that must prove compliance with privacy regulations, AI‑assisted verification could become a new way to demonstrate security without lengthy manual audits.
