# AI-Written Formal Proofs Become a Software Assurance Tool

Agent-generated, machine-checked proofs are lowering the cost of formally verifying production code, so more security-critical software will ship with proof artifacts and human effort will shift to writing and auditing specifications.

- Conviction: 38 / 100 (weakening)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-11-11T00:00:00.000Z
- Last updated: 2026-11-12T14:14:59.720Z
- Canonical: https://polylog.news/ai/trends/ai-formal-verification-of-software
- Publisher: Polylog

## Recent score history

- 2026-11-11: 40
- 2026-11-12: 38

## Recent evidence

- [confirms] DeepMind's AlphaProof Code Agents Produce Lean Proofs That Signal's Compiled Encryption Code Meets Its Specification (2026-11-11): DeepMind's AlphaProof code agents produced Lean proofs that Signal's compiled encryption code meets its specification. They generated millions of lines of machine-checked proof in about four weeks against roughly 26,000 lines of human-written spec, and the checks correctly failed on two reintroduced bugs. This is production, security-critical code getting proof artifacts, with human effort concentrated on the specification, as the thesis predicts.