The end-to-end set of deny-by-default gates that make AI-authored output verifiable: source substantiation, shape-only disclosure, and per-artifact provenance.
How it works
Deny-by-default gates sit between an AI author and a shipped artifact: every claim must trace to a substantiated source, every disclosure stays at a shape level rather than reproducing the underlying method, and every artifact carries its own provenance record. A failure on any one link blocks the artifact rather than letting it through. The gates run in sequence and the chain's contract is its order: a source-substantiation pass is not a license to skip shape-only, and shape-only is not a license to skip provenance.
Why it matters
The chain validates compliance with its gates, not the truth of its content. A piece that clears every link can still be wrong on a dimension no gate checks: structural gates pass on rules they can articulate, and an editorial gate (a manual sweep) admits exceptions a structural gate would not. Every link also adds write-time friction the reader may never inspect: the trade is author velocity for trust evidence that earns its place only when a reader needs to verify. The chain raises the floor; it does not lift the ceiling.
In practice
An AI-authored page is held at sequential gates. A claim without a cited source fails the substantiation gate. Prose that reproduces a method below the shape line fails the disclosure gate. An artifact missing its provenance block fails the provenance gate. Missing any one of the gates blocks publication; passing every gate says the page met the chain's contract, not that the page is right on every fact the gates do not cover.
Related standards and prior art
- NIST AI 600-1 (Generative AI Profile) · 2024-07-26 · (canonical framework) addresses content provenance for generative AI; the authoring-verification framing here is distinct
Defined by Ready Solutions AI