# Formal Verification Becomes the Trust Layer for AI Output

As models generate more mathematics and code than humans can review, machine-checkable artifacts become the accepted proof of correctness, and demand shifts toward domains where an automated checker exists.

- Conviction: 37 / 100 (weakening)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-09-09T00:00:00.000Z
- Last updated: 2026-09-14T14:04:09.672Z
- Canonical: https://polylog.news/ai/trends/formal-verification-as-ai-trust-layer
- Publisher: Polylog
- Affected regions: Global

## Recent score history

- 2026-09-13: 39
- 2026-09-14: 37

## Recent evidence

- [confirms] Clay Institute Declines to Call Navier-Stokes Solved After OpenAI's AI Proof Claim (2026-09-12): OpenAI says an internal model produced a claimed finite-time blowup proof in 88 hours and formalized it in Lean, making the machine-checkable artifact — not the prose argument — the thing reviewers can actually verify. The Clay Institute's refusal to credit it turns on scope (an external forcing term outside the prize problem), not on the Lean check, showing formalization settles correctness while leaving problem-statement fidelity as the remaining human judgment.
- [confirms] Anthropic Says Claude Produced the First Machine-Checked Proof of Fermat's Last Theorem in Eleven Days (2026-09-09): Anthropic says a Claude run produced the first machine-checked proof of Fermat's Last Theorem in eleven days, writing 13 million lines of Lean and about 29,500 supporting theorems against a funded human project scheduled through 2029. Unlike the contested Navier-Stokes claim, the output is a checkable Lean artifact, which is exactly the shift the thesis predicts: correctness gets settled by an automated checker rather than by peer reading.

1 more evidence entry, the full score history, the conviction-driver timeline, and affected assets are for subscribers: https://polylog.news/pricing