The Non-Human Identity Reckoning — and Why Agents Make It Urgent

For twenty years, identity security meant securing people: provisioning employees, enforcing MFA, reviewing access, offboarding leavers. We got good at it. But while we perfected the human identity lifecycle, a second population quietly took over the enterprise — service accounts, API keys, OAuth apps, service principals, workload identities, CI/CD tokens.

In most organizations these non-human identities (NHIs) now outnumber human ones by 82 to 1 (CyberArk, 2025), and the population grew 44% in a single year (Entro Labs, 2025). The governance we apply to them is a fraction of what we demand for a single employee. No consistent owner. No least privilege. Static secrets that were minted years ago and never rotated. No offboarding when the workload they served is long gone.

That was already the enterprise’s largest ungoverned attack surface. Then AI agents arrived — and made it urgent.

An agent is an identity that acts on its own

An AI agent is a non-human identity with a new and dangerous property: autonomy. It authenticates, holds privileges, and acts — increasingly without a human approving each action. Give an autonomous agent broad scopes and a long-lived secret in production, and you’ve created a credential that can reason its way into places a static service account never could. The identity questions don’t change — who owns it, what can it access, how does it authenticate, when is it revoked — but the blast radius does.

We built our identity programs around the assumption that a person sits behind the account, a manager attests to access, and a joiner-mover-leaver rhythm keeps it honest. NHIs and agents break every one of those assumptions.

Why our current programs don’t cover it

Human-identity governance assumes a person to hold accountable. Secret scanners tell you how many credentials you have; they don’t tell you which ones are crown jewels or what to fix first. And almost nothing on the market yet treats an autonomous agent as what it is: a privileged identity that needs a tier, an owner, least privilege, short-lived credentials, and a kill switch.

The fix: identity discipline, pushed to its conclusion

The answer isn’t exotic. It’s the discipline we already know, extended to a new population:

• Inventory everything non-human, agents included, and give each one an accountable owner.
• Risk-tier by posture, not by type — privilege, credential longevity, exposure, and autonomy decide what’s a crown jewel.
• Kill long-lived secrets. Move to workload identity federation and short-lived, auto-issued credentials.
• Enforce least privilege ruthlessly — no wildcard scopes, no standing admin, especially for agents.
• Close the offboarding gap. An NHI unused for 90 days is a liability, not an asset.
• Govern agents as first-class identities — inventory, least privilege, lifecycle, and continuous monitoring, mapped to the OWASP NHI Top 10 and NIST AI RMF.

This is the work I do, and I build tools to make it real rather than aspirational. I’ve written up the full control model in A Control Framework for Non-Human & Agentic Identity, and released an open-source risk-tiering scanner for non-human and agent identities, nhi-scan.

Because the organizations that get ahead of this won’t be the ones with the best policy deck. They’ll be the ones who inventoried every identity — human or not — and could answer, for each: who owns you, what can you reach, and how do I turn you off?