Episode 200 - GluuFederation/identerati-office-hours GitHub Wiki

Title: Why Agentic AI Forces a Rethink of Least Privilege

Channels

Description

AI agents don’t follow static workflows, which means traditional “least privilege” models quickly collapse into overpermissioned chaos. This IOH will explore why task-scoped authorization is essential for agentic systems, and what role AI Identity Gateways can play by implementing least-privilege by minting new tokens based on context, intent, and risk.

Homework

Takeaways

  • ⚡ Least privilege is evidence you're doing the right thing. It's not a strategy. Traditional IGA assumes long-lived identities and standing permissions. Agentic AI flips that model: ephemeral agents may exist for seconds, making it impossible to govern them with human approval workflows. The real control point becomes what agents can do—the capabilities, APIs, tools, and data and resources they can access.

  • ⚡ Agentic AI demands a shift from standing permissions to just-in-time authorization. Fine-grained, short-lived permissions (such as tokens scoped to a single tool invocation and lasting only seconds) are a more effective defense than broad, persistent entitlements.

  • ⚡ Visibility is the prerequisite for governance. Whether it's cloud IAM policies, AI model guardrails, or agent behavior, organizations cannot govern what they cannot see. The move toward opaque AI systems creates a tension between safety guardrails and the transparency needed for risk management and accountability.

  • ⚡ The future of governance is capability-centric. AI exposes the limits of identity-centric governance. While identities may become transient, the enterprise capabilities they invoke—such as querying a database, invoking an MCP tool, or modifying a business resource—remain finite, measurable, and governable.

Livestream Archive