DeepMind's AlphaProof Code Formally Verifies Signal's Production Code in Four Weeks
On November 10, Google DeepMind released AlphaProof Code, a formal specification and verification framework built from multiple LLM agents, and announced a milestone: formal verification, at the machine-code level, of Signal's real production cryptographic implementation—the first time AI-driven mathematics has proven key security properties of the actual code of this widely deployed encryption protocol. According to @swarat (a DeepMind team member), the verification was completed in about four weeks, and the team has released the paper and Lean proof code.
Confirmed
- AlphaProof Code is a multi-agent framework for deep formal verification, with the Signal protocol's core cryptographic code as its first project
- The verification was performed at the machine-code level, covering key security properties of the real implementation
- @TortainedPoet30 (original author name @TorturedPoet30) relayed that the system is powered by Gemini Carbon and generated over 6 million lines of related content
- The paper and Lean proof code have both been made public
Why It Matters
- This is the first time AI has completed formal verification at this depth on production code of a widely deployed cryptographic protocol; security community figures like @jedisct1 see it as a major breakthrough for software engineering and cybersecurity
- Formal verification has previously been so costly that verifying production cryptographic code within a reasonable timeframe was nearly impossible; AI agents could dramatically lower that barrier
2026-11-10 ~ 2026-11-10 · 6 related posts
Primary sources
- DeepMind's AlphaProof Code Formally Verifies Signal Protocol in a Month — sebkrier · 2026-11-10
- DeepMind's AlphaProof Code Formally Verifies Signal Crypto in Four Weeks — TorturedPoet30 · 2026-11-10
- DeepMind's AlphaProof Code proves Signal's compiled code secure in 4 weeks, humans review only 1% — javimbuilds · 2026-11-10