Navigating Autonomy: The Intersection of Agents and Governance in the Data Layer

Empowering AI Agents: The Importance of Governance at the Data Layer

Presented by EDB

As artificial intelligence (AI) agents gain more autonomy in decision-making, a critical question arises: What mechanisms are in place to prevent unauthorized actions by these agents? This question is at the forefront of architectural considerations for enterprises utilizing AI technology.

When AI agents have the ability to act independently across systems without human intervention, the responsibility for their actions falls on the organization deploying them. Reactive measures or abstract policies on paper are insufficient to address the real-time decision-making capabilities of AI agents. Instead, intelligent rules that consider the context of each situation are essential.

For instance, a simple rule such as “Never open the car door” may seem straightforward, but in a crisis situation like a car accident, the rule may need to be overridden for safety reasons. Context is key in guiding AI agents to make informed decisions.

While traditional governance approaches involve adding guardrails and monitoring mechanisms around AI agents, these methods have limitations, especially when agents operate autonomously and in milliseconds across multiple systems simultaneously. Governance needs to be executable and enforced at the operational data layer where AI agents interact with data in real-time.

The Role of the Data Layer in Governance

AI agents derive value from accessing and manipulating data. Policies that restrict access to certain data must be enforced at the moment the agent requests it. Additionally, auditability of AI actions requires a comprehensive record of data interactions, user identities, and outcomes. By implementing governance at the data layer, organizations can ensure compliance regardless of the agent’s behavior.

Enforcing Policies in a Probabilistic Environment

Enterprises should not rely on AI models to voluntarily comply with policies. Instead, policies must be enforced at the system level to prevent unauthorized actions. This proactive approach ensures that AI agents operate within predefined boundaries set by the organization.

Key controls at the data layer include role-based access control, data masking, audit logging, and encryption. By treating AI agents as distinct entities with declared purposes, organizations can evaluate their actions in real-time and maintain a record of their activities.

Key Imperatives for Governance

Enforce it

• Role- and attribute-based access control at query time
• Dynamic column masking based on policies
• Agent identity as a primary entity with declared purpose

See it and prove it

• Data classification and tagging for policy enforcement
• Session-level audit logging for tracking agent activities
• Lineage tracking across data pipelines

Unify and harden

• Centralized policy management
• Data encryption at rest and in transit
• Consistent enforcement across different environments

By integrating declared purposes into access control mechanisms, organizations can enhance governance and accountability for AI agents’ actions. This approach enables faster adoption of AI technologies while maintaining security and compliance standards.

Striking a Balance: Leveraging AI Capabilities Safely

The goal of governance at the data layer is not to inhibit AI agents from performing tasks but to define clear boundaries for their actions. By implementing robust governance practices, organizations can monitor, audit, and control AI agents effectively, instilling trust in their operational models.

Embracing Openness and Sovereignty in Data Governance

EDB’s open-source Postgres platform offers enterprises a secure and transparent foundation for managing AI workloads. By prioritizing data sovereignty and enforcing governance at the data source, organizations can ensure compliance with regulatory requirements and maintain control over their data.

As AI technologies advance, organizations must proactively establish governance frameworks that align with their operational models. By enforcing governance at the data layer, enterprises can confidently leverage AI capabilities while safeguarding their data assets.

For more information on governing AI in the enterprise, refer to EDB’s white paper Governing Agentic AI at Enterprise Speed.

Max Romanenko, Chief Technology Officer at EDB

Disclaimer: This article is a sponsored content piece. For inquiries, contact sales@venturebeat.com.