Picture a company that deploys an AI agent to help employees find and reserve meeting rooms. Its intended job is simple: Check room availability and book an appropriate space. But what is it capable of doing? That’s a different question.
If an AI agent’s permissions are too broad, one designed to reserve rooms could end up reading sensitive calendar details or even changing meetings it doesn’t own. Without clear AI agent ownership, no one may notice, or know whom to call, when that happens.
In my conversations with business leaders, I often hear about AI agent fleets numbering in the thousands or more, many without a listed owner. So what happens when an AI agent that was supposed to book a meeting room logs in to your email account instead?
When agentic AI became more popular, I saw many organizations go in one of two directions. Some rushed to deploy as many AI agents as possible to keep pace with competitors. Others banned agents outright, which in many cases pushed use underground. According to our research, 29% of security leaders have discovered AI agents in production in the past year that they didn’t know existed.
The rush and the panic often led to the same place: Ownership wasn’t a required field from day 1. Fewer than half of organizations (48%) have a governance framework that covers AI agent ownership, intended use, and permissions, our research found.
Many leaders may also underestimate how far behind they are. More than half rate their agentic security as “established,” meaning they have demonstrable, purpose-built governance. Yet only 20% have a single, named owner accountable for agentic AI governance, and just 13% have a defensible posture.
We’ve seen a milder version of this story play out before. During the software as a service (SaaS) boom, shadow IT became a serious threat to workplace and customer security. Security and risk teams responded by investing heavily in SaaS Security Posture Management (SSPM) and Cloud Infrastructure Entitlement Management (CIEM) to strengthen accountability for applications.
Rogue AI agents can be harder to contain. Traditional software tends to behave predictably. An AI agent acts autonomously, and its response to a given prompt can be difficult to anticipate.
Still, two lessons carry over from the SaaS world:
- Build governance into the architecture from day 1. Organizations that add security later often find themselves pinning rules and permissions to a moving target. Yet only 44% build in security before development, according to our research.
- Make accountability clear. Assigning each AI agent a named owner, whether an individual or a designated role, can help organizations respond faster when an agent behaves unexpectedly.
Standards bodies increasingly point in the same direction. NIST’s AI Risk Management Framework calls for roles and responsibilities related to managing AI risks to be documented and clear to individuals and teams throughout the organization.
In February 2026, NIST’s National Cybersecurity Center of Excellence released a concept paper arguing that enterprises need to treat AI agents as identifiable entities within their identity systems, with their own credentials.
In October, NIST published an initial set of design tenets for AI agent identity, authentication, and authorization. One calls for agents to stay linked to the people or organizations that task them throughout execution. Without that link, NIST warns, organizations face severe nonrepudiation problems, where high-impact actions can’t be attributed to anyone.
Consider the moment an AI agent misbehaves. Someone has to decide whether to pause it, who joins the investigation, and who approves the fix. With no clear owner, those decisions can bounce between IT, security, and the business while the agent keeps running.
A named owner can simplify that response. It’s the owner’s job to understand what the AI agent was built to do, help coordinate revoking its access, and help investigators understand its intended purpose.
Ownership works alongside an organization’s broader security and governance controls, giving them a clear point of contact. It may also change behavior before anything breaks: A person who knows their name is attached to an AI agent may be more likely to ask what it can reach before they deploy it.
Naming an owner could be the difference between an AI agent that manages your calendar and one that reads your email.
Find out how ServiceNow can help you assign an owner for every AI agent across your business.