10 to the 23 AI logo

Practice Library

Governance patternstructural

Provenance labeling

Stamp unverified and AI-generated records so people and systems down-weight them instead of inheriting them as fact.

What it changes

dampenedContaminated records read and believed
dampenedContaminated records repeated as fresh output

Who can pull it

Deploying organizationHarness builderData-protection officer

What it looks like institutionally

A record system that cannot distinguish verified fact from unverified draft treats both as truth. Provenance labeling makes the distinction machine- and human-readable: AI-drafted, human-verified, source-linked, stale-since. Readers calibrate; retrieval systems filter; audits target.

This is the cheapest intervention on the record-to-people and record-to-model pathways, because it changes how contamination behaves without having to find it first: unverified material stops spreading at full credibility even before anyone cleans it.

The implementation detail that matters: labels must survive copying. A provenance stamp that vanishes when a paragraph is pasted into a new assessment governs nothing.

Addresses: Contaminated records read as fact · Model retrieving its own errors. Test a version of this lever in the PAN Lab.

Deciding whether this lever fits your deployment?

Which patterns matter — and in what order — depends on your system's actual shape. Ranking your options on evidence, with what can backfire stated, is engagement work.