When agents act on their own, governance has to live in the data layer

The article argues that as enterprises give AI agents more autonomy, traditional guardrails such as instructions, policies, and monitoring are not enough to prevent unauthorized actions. It says governance must become executable and enforced at the operational data layer, where agents actually query, transform, and act on enterprise data.

The piece reflects a broader shift from model-level AI safety toward infrastructure-level controls that can operate in real time. As agents gain more ability to act independently, enterprises may increasingly treat the data layer as the central point for access control, policy enforcement, and auditability.