DeepMind's AlphaProof Code proves Signal's compiled code secure in 4 weeks, humans review only 1%
javimbuilds · reddit · 2026-11-10
Google DeepMind's AlphaProof Code produced machine-checked security proofs for Signal's actual compiled code — not just the protocol spec — in 4 weeks:
- Gemini-powered agents wrote over 6 million lines of Lean proofs, verified automatically by Lean
- Human experts reviewed only the 1% the agents flagged
- Verification covers the implementation layer, where most protocol efforts stop — and where real bugs live, in the gap between spec and code
- As a sanity check, the team reintroduced two old, fixed Signal bugs, and the system generated proofs confirming those versions were insecure
The Reddit poster's key caveat: humans still wrote and reviewed every spec, so the proof is only as good as what you chose to prove — writing correct specs may remain the real bottleneck.
More from coding & agent
- Adobe Firefly Builder (beta): describe a website, get a live editable React app — rufusd · 2026-11-11
- Zero-dependency MCP hub fronts nine research repos as 14 tools without merging dependency trees — Emergency-Tutor-7 · 2026-11-11
- Video Decompiler: Open-Source Skill Has Fable 5.2 Code Every Frame of a Video — brandon_galang · 2026-11-11
- Gemini 4 Carbon tested: 100+ unattended passes rebuild a 3D fort in a raw engine — Prompt Engineering · 2026-11-11
- 53% Less Code: Ponytail Forces Agents to Ask 'Does This Need to Exist?' Before Writing — nevrekaraishwa2 · 2026-11-11
- YC F26 Startup Elimentary Launches AI Agents for Private Credit Ops, Cuts NAV Checks to 30 Minutes — ycombinator · 2026-11-11