Enforcing Fine-Grained Security at Machine Speed: Dynamic Access Control for High-Frequency AI Agents

This article addresses the security challenges posed by high-frequency AI agents in analytics, contrasting them with human users. It argues against using broad service accounts, advocating for identity propagation, dynamic access control, and policies evaluated at request time. Key topics include row/column/metric permissions, short-lived credentials, audit trails, rate limits, and semantic policies. The article emphasizes governed lakehouse access (e.g., Dremio) to ensure agents operate with proper scoping, delegation, and accountability, preventing super-user token abuse and unintended data disclosure. It provides a practical checklist for implementing secure agentic analytics.