Identity Management Day: How AI Agents are Redefining Identity Security

As the cybersecurity community continues to focus on identity management and awareness for Identity Management Day, the conversation around securing digital identities is undergoing a major shift.

For years, identity security has been broadly categorised into two buckets: human identities and non-human identities (NHIs). However, the rapid emergence of AI agents is blurring these lines, forcing security professionals to rethink how they manage access and permissions.

As we look toward the future of identity security, it is crucial to understand how AI agents are poised to disrupt the non-human identity space and what organisations must do to prepare.

A non-human identity (NHI) is any credential, access token or digital persona used by a system, application, device or machine to authenticate and interact with other systems.

Long before the AI boom, networks were already teeming with NHIs. Common examples of NHIs include:

• Service accounts used by applications to run automated processes or access databases
• Application programming interface (API) keys used by software to communicate with third-party cloud services
• Certificates used to authenticate devices, servers and workloads
• Bots and scripts: automated tools executing predefined tasks within IT environments

According to a study by Silverfort, NHIs outnumber human identities by a ratio of 50 to one. 

Managing these traditional NHIs has never been easy. Unlike employees, who have centralised HR processes for onboarding and offboarding, NHIs are frequently created by developers on the fly, undocumented and sometimes left active long after they are needed.

For instance, many organisations struggle with API keys and credentials that end up hardcoded into public repositories or with over-privileging – when service accounts are granted broad administrative rights simply to ensure a script doesn't break.

Organisations attempt to manage these risks using privileged access management (PAM) solutions, secrets vaults and automated lifecycle management tools to rotate keys and enforce least privilege.

However, traditional strategies rely heavily on static rules and predictable machine behaviours. AI agents could fundamentally disrupt that predictability.

Traditional NHIs are deterministic. A service account executes a specific, hardcoded task. An API key accesses a specific endpoint. AI agents, however, are autonomous and goal oriented.

Agentic AI refers to systems capable of understanding a high-level goal, breaking it down into steps and independently executing a series of actions across various applications to achieve that goal.

For example, an AI agent might be tasked with "optimising the supply chain software." To do this, the agent might need to read emails, query a secure database, communicate with a vendor's API and write new code, performing all these tasks autonomously and in real time. In the future, several agents may perform tasks simultaneously.

This introduces a completely new type of dynamic digital entities acting on behalf of humans but making independent operational decisions.

In an interview on The Verge’s Decoder podcast, Okta’s co-founder and CEO, Todd McKinnon, highlighted this exact paradigm shift.

McKinnon argued that AI agents represent a completely new category of identity that sits squarely between human identities and traditional NHIs.

Because these agents act on behalf of a specific human user – inheriting their intent and potentially their permissions – but execute tasks with the speed and autonomy of a machine, they cannot be neatly placed into our existing identity buckets. 

This new class of identity brings unprecedented challenges for cybersecurity and identity management.

How do you enforce "least privilege" on an AI agent whose required actions change dynamically based on the problem it is trying to solve? If an agent requires access to a financial system on Tuesday and a source code repository on Wednesday, static role-based access control (RBAC) models will either block the agent from doing its job or require granting it dangerous, overly broad permissions.

When an AI agent interacts with a software-as-a-service (SaaS) application on behalf of a human, how does the system know if the action was initiated by the employee or independently hallucinated by the agent? The line between legitimate delegation and dangerous autonomous action becomes incredibly thin. If an AI agent falls victim to a prompt injection attack, it could be hijacked to exfiltrate data while wearing the authorised identity of its human owner.

If an AI agent creates its own sub-agents to complete micro-tasks, who is tracking the lifecycle of those temporary identities? Security teams will need continuous, contextual authentication mechanisms that monitor how an identity is behaving in real-time, rather than just checking if it has the right cryptographic key at the door.

The disruption of the non-human identity space by AI agents is no longer a theoretical exercise, but an impending reality.

To secure this new frontier, organisations must evolve beyond static secrets management. Security teams will need to embrace continuous identity posture management, highly granular just-in-time (JIT) access, and behavioural analytics that can spot when an agentic identity starts acting outside its expected parameters.

Identity Management Day serves as a reminder for the cybersecurity community to stop viewing identity security strictly through the lens of human employees and static service accounts.

The AI agents are here, they are (almost) autonomous, and they need identities. It is up to cybersecurity professionals to ensure they are secure.

ADVERTISEMENT

Enjoyed this article? Make sure to share it!

Looking for something else?