The frequently cited ratio of non-human identities to human users captures a reality that security teams can no longer treat as background infrastructure. Service accounts, certificates, workloads and automation already outnumber people in many environments. AI agents now increase the pace, autonomy and complexity of that identity population.
The problem
The challenge is not simply volume. Machine identities often lack a consistent naming standard, a human owner or a clear expiry date. They can be duplicated across environments and granted broad permissions because developers need systems to work quickly. When an agent is added to that environment, it may combine several inherited privileges and make decisions that were previously performed manually.
What security teams should prioritise
A workable response starts with identity-level inventory. Organisations need to map each agent and workload to its credentials, parent application, owner, purpose and reachable resources. That graph should be updated as code changes and deployments occur, not only during periodic reviews.
Least privilege must also become dynamic. An agent that normally reads ticket data should not suddenly create privileged cloud roles without a policy decision and strong telemetry. Context-aware controls can limit actions by workload, time, destination and sensitivity, while just-in-time access reduces standing exposure.
Governance teams should define accountability before scaling agentic automation. Policies need to cover instruction changes, tool registration, credential issuance, emergency revocation and evidence retention. Treating agents as first-class identities makes NHI security measurable and gives incident responders a way to reconstruct what happened.