Agentic AI in Cybersecurity: Governance and Security

Agentic AI in Cybersecurity: What It Means, Why It Matters, and How to Govern It

Key Takeaways

• AI agents create non-human identities that require the same visibility, governance, and accountability as human users.
• Unmanaged AI agent permissions can expand attack paths and increase the blast radius of a security incident.
• Gartner predicts 33% of enterprise applications will include agentic AI by 2028, up from less than 1% in 2024, creating substantial identity governance challenges.
• Effective AI governance starts with discovering identities, understanding permissions, and enforcing lifecycle controls.

What is agentic AI in cybersecurity, and why does it matter?

Agentic AI is creating a new identity security challenge. AI agents can independently access systems, use credentials, and take actions across the enterprise. As these agents become more autonomous, organizations must govern them as non-human identities to maintain visibility, control, and accountability.

• Every AI agent functions as a non-human identity.
• AI agents often operate through API tokens, OAuth permissions, service accounts, and cloud roles.
• Without governance, AI agents can become shadow AI assets operating outside security oversight.
• Identity security provides the foundation for governing AI agent access and behavior.

Quick Facts

Agentic AI Security Risks and How to Address Them

The AI Access Governance Gap is Growing

Organizations are rapidly deploying AI agents, but only 21% of enterprises have mature governance models in place. That oversight leaves organizations lacking the visibility and identity controls needed to manage AI agents. This results in a rising tide of unmanaged non-human identities with unknown permissions, credentials, and access paths.

Consider a security team that deploys an AI agent to automate cloud cost optimization. To do its job, the agent receives access to AWS resources, GitHub repositories, Kubernetes clusters, and ticketing systems through API tokens, OAuth permissions, and service accounts.

Initially, the agent saves several hours of manual work each week. Over time, however, additional integrations are added. New permissions are granted. Temporary access becomes permanent. Six months later, no one can confidently answer a few basic questions:

• What systems can the agent access?
• Which credentials is it using?
• Who owns it?
• Are its permissions still necessary?
• What would happen if the agent were compromised?

This scenario is becoming increasingly common. As AI adoption grows, security teams are discovering that traditional IAM programs often lack visibility into these identities. OWASP recently introduced the OWASP Non-Human Identity Top 10 to address growing security concerns around machine identities, service accounts, API keys, and AI-driven workloads.

Why Does Agentic AI Create New Security Challenges?

Agentic AI creates new security challenges because every AI agent requires access to systems, data, and resources. That access is typically granted through API tokens, OAuth permissions, cloud roles, service accounts, secrets, and other machine identities.

As AI agents become more autonomous, they increasingly interact with multiple systems, expanding both their capabilities and their access. The challenge is no longer simply identifying AI agents. Effective AI access governance requires understanding the permissions, credentials, and access relationships that enable them to operate.

This reinforces a fundamental reality: AI security is ultimately an identity security challenge. Maintaining visibility, control, and accountability requires organizations to govern the identities, permissions, and credentials behind every AI agent.

Consider our cloud optimization example. The agent began with limited permissions, but as new integrations were added, it accumulated access to source code repositories, cloud resources, and operational systems. This type of permission growth is common and often difficult to track. Without continuous governance, security teams may not realize how much access an AI agent has accumulated until after an incident occurs.

Why AI Governance Cannot Be an Afterthought

Neglecting AI governance creates operational, security, compliance, and business consequences. A compromised AI agent can act far faster than a compromised human account because it is designed to automate actions at scale. The impact can include:

• Unauthorized access to sensitive data
• Excessive cloud resource manipulation
• Exposure of API credentials
• Compliance violations
• Incident investigation challenges
• Expanded attack paths across environments

However, Deloitte reports that only 21% of enterprises responding to a recent multicountry survey report having mature governance in place to manage the risks of agentic AI.

Imagine our AI agent is compromised through a leaked API token. Because the agent has access to cloud infrastructure, repositories, and ticketing systems, an attacker could use a single identity to move across multiple environments.

A useful example of the critical nature of governance comes from the 2023 Okta support system breach. Attackers leveraged access through support-related processes and identity infrastructure to gain visibility into customer environments. While not caused by AI agents, the incident highlights a core lesson: access relationships matter. Organizations that cannot map identity relationships often struggle to understand exposure during an incident.

How Would Leading Security Teams Address Agentic AI Security?

Savvy security teams treat AI agents as first-class identities, extending identity governance beyond human users to include machine identities, service accounts, OAuth applications, and AI agents. Regardless of the identity type, effective governance follows the same three-step framework:

• Discover: Identify AI agents, service accounts, OAuth applications, API tokens, machine identities, and their owners.
• Understand: Analyze permissions, credential usage, access relationships, and potential blast radius to determine the risks associated with each identity.
• Enforce: Apply least-privilege policies, credential rotation, lifecycle governance, ownership validation, and automated remediation to reduce risk over time.

This context is critical because visibility alone does not reveal risk. Security teams must understand how identities interact, what systems they can reach, and how a compromise could affect the broader environment.

Following our example, the security team would not treat the cloud optimization agent as a standalone application. Instead, they would govern it as a non-human identity. They would discover the agent and its associated credentials, understand its permissions across cloud, development, and operational systems, and enforce controls to ensure its access remains appropriate.

How Token Security Approaches Agentic AI in Cybersecurity

The Challenge

AI agents create a rapidly growing population of non-human identities. Security teams often struggle to determine where these agents exist, what permissions they possess, and how those permissions create exposure across cloud, SaaS, and development environments.

In our example, the security team must identify the cloud optimization agent, understand its access to connected systems, and determine whether its permissions remain appropriate.

The Approach

Token Security applies a machine-first security model focused on non-human identities and AI agents. Rather than adapting human identity controls to machine-driven environments, it helps organizations govern the identities, permissions, and credentials that enable AI agents to operate.

• Discover AI agents, machine identities, service accounts, and associated credentials through identity discovery.
• Understand permissions, access relationships, and potential blast radius through entitlement mapping, behavioral baselines, and risk analysis.
• Enforce governance controls through lifecycle governance and automated remediation.

The Outcome

Organizations gain greater visibility into AI agents and machine identities, enabling security teams to establish clear ownership, understand access risk, reduce excessive permissions, and remediate identity-related findings more quickly. Visibility alone is not enough. Effective governance also requires continuously enforcing least-privilege access, credential hygiene, and lifecycle controls as AI agents evolve.

From Shadow AI to AI Access Governance: How Security Teams Are Gaining Control

HiBob Gains Visibility into Shadow AI

As organizations experiment with custom GPTs and AI agents, security teams often struggle to determine which agents exist and what access they possess.

"Token Security gives us visibility we simply didn't have before. We can now automatically identify and control custom GPT agents running in our environment and ensure the required security level. Knowing that no AI agent is operating beyond our oversight means we can confidently accelerate our AI adoption."

This example highlights a common governance challenge. AI adoption often moves faster than security visibility. Identifying and governing AI agents becomes essential for controlling shadow AI.

Lemonade Improves Machine Identity Risk Visibility

Security teams frequently receive large volumes of alerts but little context about which findings matter most.

"Token Security has enhanced our security by providing us accurate and relevant visibility into machine identity risks, something I have yet to see from other vendors. It's the first service I've seen that delivers on the widely-made claim of finding machine identity risk. It provides us with the right amount of detail to mitigate issues without burdening us with tens of rows of useless alerts."

This reinforces an important lesson: visibility alone is not enough. Security teams need context, ownership, and remediation guidance to reduce machine identity risk effectively.

Why Identity is the Control Plane for Agentic AI

AI agents are becoming operational participants across the enterprise, writing code, querying data, managing infrastructure, and automating workflows.

Consider the cloud optimization agent in our example. What began as a productivity initiative evolved into a non-human identity with access to critical systems. As agentic AI adoption accelerates, that progression is becoming increasingly common.

Every AI agent operates through identity. As a result, securing agentic AI requires more than securing applications. It requires governing the AI agents, service accounts, API tokens, and machine identities that enable access.

Identity security is the control plane for agentic AI, enabling organizations to understand access, limit blast radius, and maintain accountability at scale. Organizations that can discover, understand, and govern these non-human identities will be best positioned to scale AI securely.