Research questionHow can autonomous research swarms govern shared channels without amplifying evaluation exploits?Shared knowledge stores and agent-to-agent channels can let an evaluation exploit spread through a swarm before operators detect or contain it. The same visibility can support fraud detection, whistleblowing, and collective enforcement, making containment a governance problem rather than a simple communication shutdown.