Is Your Identity Framework Ready for Agentic AI?

AI agents are proliferating—autonomously booking meetings, analyzing data, making recommendations, and increasingly, taking actions on behalf of users. But the truth is that the identity and access management (IAM) systems organizations have relied on for decades weren't built for Agentic AI.

DOWNLOAD NOW: Accelerate Secure Adoption of Agentic AI

What is the Third Identity Problem?

For years, security teams have operated with a simple worldview: human identities and machine identities. Humans are slow, context-aware, and unpredictable. Machines are fast, deterministic, and predictable. Our IAM systems, protocols like OAuth and SAML, and our entire control frameworks reflect this division. Agentic AI breaks this model.

AI agents combine the speed and scale of machines with the creative, less predictable behavior of humans. They can be manipulated like people through prompt injection or social engineering, yet they operate at machine speed and scale. This hybrid nature creates a gap: they move too fast for human-centric controls and too unpredictably for machine-centric ones.

What results is an explosion of unmanaged identities, privileges that shouldn't exist, and an accountability vacuum that should worry any CISO thinking about the next audit.

READ MORE: Tips to Accelerate the Secure Adoption of Agentic AI

Why Legacy IAM Falls Short

Traditional IAM was designed for long-lived accounts with static, coarse-grained permissions. But agentic AI demands something fundamentally different. There's a path forward, anchored in the NIST AI Risk Management Framework and zero trust principles. Here's what it looks like in practice:

1. Adopt Just-in-Time Everything

The entire lifecycle of an AI identity must be automated. Agents should be provisioned at runtime for specific tasks and deprovisioned as soon as they're complete. That means there can be no standing privileges, no static API keys, and no accounts that outlive their purpose. This is the only way to manage the scale and velocity of agentic AI without creating too many credentials.

2. Implement Persona Shadowing

Give each agent its own unique identity that “shadows” the delegating human. This creates an unambiguous audit trail linking every agent action back to the person who authorized it.

3. Build Delegation Chains

In multi-agent workflows, resist the temptation to pass broad permissions down the chain. Each delegation step should create a new, more narrowly scoped credential. This preserves least privilege and contains the blast radius when an agent is compromised.

4. Monitor Behavior, Not Just Signatures

Implement comprehensive, real-time monitoring with an immutable audit trail. Logs must contain rich, structured metadata, including the agent ID, delegating user ID, and task ID. For high-risk operations, implement human-in-the-loop verification—requiring explicit human consent before execution.

READ MORE: How to Effectively Use AI

The Shadow AI Problem

While you're building governance frameworks, shadow AI is already in your organization. Employees are uploading sensitive data to ChatGPT, Claude, and dozens of other AI tools in an effort to increase productivity. The most effective way to eliminate shadow AI is to address why it exists: unmet business needs. A smart strategy is to understand what end users need for AI tools and create options for them to adopt:

• Discover: Use CASBs, SMPs, and EDR to make shadow AI visible
• Educate: Launch continuous awareness about risks and approved alternatives
• Provide alternatives: Give employees powerful, secure, vetted AI tools

Begin Your Agentic AI Journey Now

Agentic AI is here. Organizations must determine how they will govern AI identities. Ideally, you will do it proactively with a governance framework—and not reactively after a breach.

The organizations that treat agentic AI as a distinct identity class, anchor their approach in established frameworks like NIST AI RMF, and invest in modern, automated IAM infrastructure, will be well-positioned to harness AI’s power securely.

DOWNLOAD NOW: The Security, Privacy, and Compliance Implications of Agentic AI

Get the Latest Analysis on the CISO Talent Landscape

You can also download our 2025 Security Organizational Design Benchmark Report—and gain access to valuable insights on team design, leadership positions, and pay ranges broken out by three distinct revenue and staffing clusters: contact us to request the full report.

Although reasonable efforts will be made to ensure the completeness and accuracy of the information contained in our blog posts, no liability can be accepted by IANS or our Faculty members for the results of any actions taken by individuals or firms in connection with such information, opinions, or advice.