Why does my AI agent route European health data through a US server without warning?

Multi-step AI workflows route personal data through agent chains, tool calls, embedding models, and LLM inference endpoints spread across cloud regions and jurisdictions. No orchestration layer lets you tag data with residency constraints and enforce them at runtime across the full pipeline. A piece of EU personal data can silently cross a jurisdictional boundary inside a single tool call, as a European insurance company discovered when fined in May 2025 after its claims processing agent routed medical data through a US LLM provider for four months without adequate transfer mechanisms. GDPR and the EU AI Act, which reaches full enforcement for high-risk systems in August 2026, create hard legal obligations, but there is no technical enforcement primitive that sits between the application and the infrastructure to verify compliance in real time.