Q3. Does this affect runtime behavior? Or is it only used at build and test time?

Both stages matter. At build time, regulatory context shapes agent behavior definitions and eval datasets. At runtime, we continuously monitor the agent's performance and regulatory compliance, which is what liability management actually requires: proof that the agent stayed compliant while it was running, not just that it passed a test once. We're also building toward real-time pre-action checking, catching a compliance issue before the agent acts rather than after. That's still early and not fully live today, but it's the direction the runtime layer is heading.

‍