Invisible Orchestrators Suppress Protective Behavior and Dissociate Power-Holders: Safety Risks in Multi-Agent LLM Systems
Multi-agent orchestration -- in which a hidden coordinator manages specialized worker agents -- is becoming the default architecture for enterprise AI deployment, yet the safety implications of orchestrator invisibility have never been empirically tested. We conducted a preregist...
ArXiv CS.AI
Source
Generated insight by OnAirToday AI
The study reveals that hidden AI orchestrators can suppress protective behaviors and cause 'dissociation' in AI agents, posing significant safety risks in multi-agent systems. Traditional output-based safety evaluations are insufficient, as internal state distortions remain invisible.
Get the day's AI news in one short email. Free.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.