The conversation around AI is changing. For the past few years, the focus has been on what AI can generate. Today, that focus is shifting towards what AI can actually do. Organizations are increasingly deploying AI agents that retrieve information, interact with business applications, trigger workflows and execute tasks with minimal human intervention.
As these agents become embedded in everyday business processes, they move beyond productivity tools and become active participants in business operations. That shift raises a new question: how should they be governed?
AI isn’t creating a new identity challenge
There is a common misconception that AI introduces an entirely new security problem. In reality, organizations have been managing non-human identities for years. Service accounts, APIs, integrations, automation platforms and workloads all require identities to authenticate, access systems and perform tasks. AI agents are simply the next evolution: they don’t replace existing identities, but add another layer to an identity landscape that is already becoming increasingly complex.
Like every other identity, AI agents need permissions to access applications, retrieve data or execute actions. In many cases, they don’t even operate under their own identity. Instead, they perform actions on behalf of a user by leveraging delegated permissions. That means the same questions organizations have been asking for years remain just as relevant.
Who owns this identity? Why does it need access? Which permissions has it been granted? Are those permissions still appropriate? And can every action be traced back and explained?
AI does not change these questions. It simply means organizations will have to answer them for a growing number of identities operating at a much greater scale.
The real challenge is maturity
The discussion around AI often starts with technology, but in reality, it should start with identity.
Many organizations are investing heavily in AI while still struggling to understand their existing identity landscape. Human identities remain fragmented across multiple systems, service accounts accumulate permissions over time and ownership of non-human identities is not always clearly defined. Access reviews are frequently focused on employees, while automated identities receive far less attention.
One of the most common examples is role changes within an organization. Employees move into new positions, take on additional responsibilities or join new projects. New permissions are added, but existing access is not always removed. Over time, users accumulate privileges that no longer reflect their current responsibilities. The result is a growing number of so-called “super users”: identities with access that has evolved historically rather than intentionally. Most organizations recognize this pattern immediately: it is rarely caused by a single decision, but by years of manual joiner-mover-leaver processes, inconsistent role management and limited visibility into existing permissions.
Agentic AI amplifies this challenge. When an AI agent performs actions on behalf of a user, it can leverage the permissions available to that identity. If that user has accumulated excessive access over the years, the AI agent can make use of those same permissions to complete its tasks.
The bottom line? AI does not create excessive privileges, but it exposes them. And as organizations deploy more AI-driven automation, weaknesses in identity governance become increasingly visible and potentially more impactful. Before organizations ask how they will govern AI agents, they should first ask themselves a simpler question: do we actually know every identity that already has access to our environment today?
Governance and visibility go hand in hand
Organizations cannot govern what they cannot see. Building a mature identity strategy starts with understanding which identities exist, who owns them, what they can access and whether that access is still justified. Only then can governance processes ensure access remains aligned with business needs throughout the identity lifecycle.
Identity Governance & Administration (IGA) plays a central role by managing the complete identity lifecycle, from provisioning and approvals to access reviews and deprovisioning. At the same time, it provides organizations with valuable insight into identities, access rights and governance processes.
Identity Visibility & Intelligence Platforms (IVIP) complement these capabilities by providing deeper visibility into identities, permissions and relationships across increasingly complex environments. This enables organizations to understand their identity landscape from both a security and compliance perspective, even before introducing extensive automation.
Together, IGA and IVIP provide the governance and visibility needed to maintain control as the number of human, non-human and AI identities continues to grow.
AI agents will also become identities of their own
Today, many AI agents operate on behalf of users by using delegated permissions. As technology matures, organizations will increasingly deploy autonomous agents with identities and permissions of their own. These AI identities will require the same governance principles as any other non-human identity: they need clear ownership, well-defined lifecycle management, least privilege access, continuous monitoring and regular access reviews.
While standards for AI identities continue to evolve, organizations don’t need to wait for a completely new governance model. The capabilities required to manage AI identities securely are, to a large extent, already available today.
The foundation already exists
The capabilities required to manage AI securely already exist. Organizations with mature Identity Governance & Administration, strong visibility into their identity landscape and a consistent least privilege approach are already building the foundation needed to govern both today’s identities and tomorrow’s AI agents. The organizations that will succeed with AI won’t necessarily be those deploying the most AI agents. They’ll be the ones that know exactly who, or what, has access to their environment, what that access enables and how it is governed throughout its lifecycle.
