AI Agent Governance Moves to the Data Layer

As enterprises grant AI agents more autonomy – the ability to plan, decide, and act across systems without human approval for each step – the question of how to stop an unauthorized action becomes central.

Companies are responsible for their agents' actions. This responsibility cannot be met with hindsight or abstract policies. Agents require rules that are contextual and executable in the moment, as they lack independent judgment.

Rather than relying solely on guardrails around the agent, such as instructions and monitoring, governance should be embedded within the data layer where agents perform their work. A policy limiting access to a data class is meaningful only if the system can deny access at the point of request.

Enforcement at the data layer is crucial because agent behavior can be probabilistic, while governance must be deterministic. Mechanisms like role- and attribute-based access control, row- and column-level security, and classification/masking can be applied directly to data. Agent identity must be managed as a first-class principal, with a declared purpose bound at session start.

The goal is not to prevent agents from doing useful work, but to define their boundaries, what they can touch, what they can change, and how events can be reconstructed. Governance at the data layer, built on open-source foundations like PostgreSQL, gives companies control over data location and policy enforcement, which is critical for regulated industries.