Non-Human Identities Are About to Outnumber Your Employees, Is Anyone Managing Them?

I want to plant a slightly unsettling thought in your head, because it's the security story almost nobody is telling and it's coming fast. Within a couple of years, the number of non-human identities in your enterprise, AI agents, bots, service accounts, IoT devices, will dwarf the number of human ones. Every one of those agents needs to authenticate, needs permissions, needs access to systems. And right now, at most companies, nobody owns managing them. That's a problem with a very large blast radius.

Think about what an AI agent actually is from a security standpoint. It's not a chatbot. It's a privileged actor, an identity that can read data, take actions, and change the state of systems of record, often with broad access because giving it narrow access was more work during setup. Multiply that by hundreds or thousands of agents, each provisioned by a different team in a hurry, and you've built an identity sprawl that would make any CISO lose sleep. The shadow-IT problem of the 2010s was unmanaged apps. The shadow-AI problem of the late 2020s is unmanaged identities with hands.

Here's why the stakes are higher than with human accounts. A compromised human account is bounded by what one person can do at human speed. A compromised agent identity can act at machine speed, across many systems, around the clock, before anyone notices. The blast radius of one over-privileged agent that gets hijacked is enormous. And because agents are designed to act autonomously, the early warning signs a human would trip, odd login times, fumbling around, don't apply.

ServiceNow saw this coming, which is why the Zurich release included a Machine Identity Console specifically for managing non-human identities: the agents, bots, and IoT devices proliferating across the environment. The fact that this shipped alongside the tools that create all those agents is the important signal. They understand that you can't responsibly hand people the power to spin up autonomous actors without also handing them a way to govern those actors' identities.

So what should you actually do? Treat every agent as a first-class identity with its own credentials, never shared, never anonymous. Apply least privilege ruthlessly; an agent should have exactly the access its job requires and not one permission more. Maintain an inventory of every non-human identity and who owns it. And ask your vendors and your own teams the blunt question: when this agent is compromised, what can it reach, and how fast will we know? If nobody can answer that, you've found your next priority. The machines are getting identities whether you manage them or not. Choose "manage."