Compositional Policy Violations: When Step-Level Compliance Fails In Agentic AI Workflows
Agentic AI workflows pass step-level checks but violate policies through sequence interactions. Real-world risks in healthcare, finance, security. Why current g