Organizational Control Layer: Governance at the Execution Boundary of LLM Agent Systems

arXiv·medium signalXBlueskyLinkedInCopy link

This arXiv paper introduces the Organizational Control Layer (OCL), a model-agnostic governance layer that intercepts agent-generated actions before execution, applying policy enforcement and escalation in economically consequential multi-agent settings. Reported results are striking: unsafe executions drop from 88% to near-zero while valid task success rises from 12% to 96%. It's a concrete blueprint for the 'action firewall' pattern enterprises increasingly need to deploy agents with real-world side effects.