AlphaProof Code, again: agents write proofs, Lean kernel gates every output — no need to trust the AI
rohanpaul_ai · x · 2026-11-10
A follow-up repost of the same "AlphaProof Code" workflow for verifying Signal's security at the machine-code level: you never have to trust the AI.
- A spec-writer agent drafts a Lean spec with the user; reviewer agents critique, user approves.
- Unmodified Rust is compiled to an AArch64 executable loaded into Lean.
- Prover agents write proofs; a gatekeeper accepts only what Lean's kernel checks and the comparator confirms against a held-out spec copy.
Humans only trust their approved security rules and Lean's kernel — any wrong AI output is thrown out by the math checker.
More from coding & agent
- Stanford NLP course's best homework: build your own agent harness — stanfordnlp · 2026-11-12
- Anthropic details agent memory: dreaming organizes Claude's memory, wiki+search hits 61% — RLanceMartin · 2026-11-12
- Talk preview: scaling 1,000 Fable 5.2 agents to formalize Fermat's Last Theorem — RLanceMartin · 2026-11-12
- Opinion: AI infra is becoming a coding runtime as intelligence units outgrow tokens — prateekj · 2026-11-12
- Claude Docs/Slides block external MCP clients: no public API, export only — steipete · 2026-11-12
- Anthropic explains agent memory: 'dreaming' lets Claude consolidate session transcripts into notes — rseroter · 2026-11-12