Opal CEO Argues AI Agent Access Needs a “GPU”-Like Decision Engine

Summary

Opal CEO Howard Ting argues that AI agents require continuous access governance. He describes a decision engine that assesses ownership, purpose, work context and resource sensitivity before granting scoped access, denying it or escalating to a person.

Key points

• Ting argues that agents’ frequent, task-specific access requests can overwhelm human review; his illustrative estimate is up to 100,000 decisions for every human access decision.
• The proposed framework checks an agent’s owner and permissions, intended purpose and behavior, task context, and the sensitivity of the requested resource.
• The decision engine can grant task-scoped access that expires, deny a request with a reason, or escalate it to a person; the article says decisions should take seconds.
• Ting recommends starting in guided mode, with a person reviewing recommendations, before allowing automated decisions for low-risk, reversible access.
• Opal describes Paladin, part of Opal Zero, as a product implementing this approach; its capabilities and performance are presented by the vendor, not independently evaluated in the article.
• The article cites Mercari’s zero-standing-access goal and says Databricks routes time-bound and session-based access through Opal; it describes agent-access decisioning at Databricks as ongoing work.
• Ting advises organizations to identify production agents and their owners, compare agent permissions with owner permissions, check access duration against tasks, and plan for owner departures.

Vendors

Products