SCOPE: Compiling Natural-Language Policies into Neurosymbolic Predicate Programs

Researchers introduced SCOPE, a system that compiles natural-language policies into neurosymbolic predicate programs for enforcing compound, application-specific rules over LLM agent transcripts and tool-call trajectories. SCOPE targets a gap the authors identify: fine-tuned guardrail models are efficient but degrade on compound policies, while LLM-as-a-judge approaches are less reliable for multi-turn and tool-call evaluation. The work addresses policy enforcement for LLM agents as requirements shift from single-turn content moderation to compound rules spanning multi-turn transcripts.

Policy enforcement for LLM agents has shifted from single-turn content moderation to compound, application-specific policies evaluated over multi-turn transcripts and tool-call trajectories. Fine-tuned guardrail models are efficient but degrade on compound policies, and while LLM-as-a-judge approach