Are AI agents non-human identities?

Yes, when they act with their own credentials. An AI agent that calls APIs, queries databases or runs commands needs an identity and secrets, and it raises the same questions as any other workload: what can it reach, how are its secrets stored and rotated, who owns it, and is its activity recorded and reviewed.

Who should own a non-human identity?

Every NHI should have a named human or team owner who is accountable for its purpose, permissions and lifecycle. Ownership makes reviews possible and stops orphaned accounts: OWASP lists improper offboarding as the first risk in its Non-Human Identities Top 10, for identities left active after the service or person behind them has gone.

Glossary

Non-human identity (NHI)

Also called NHI, machine identity, workload identity, service identity, service account

A non-human identity (NHI) is an account or credential used by software rather than a person, such as a service account, API key, workload role, bot or certificate, that lets applications, scripts and services authenticate to other systems.

Rudratic glossary

Most identities in a modern estate do not belong to people. Backup jobs, CI/CD pipelines, microservices, monitoring agents, integrations and AI agents all sign in to something, usually with a password, key, token or certificate. These non-human identities often hold broad, long-lived access and have no one watching them, which makes them a frequent target.

What counts as a non-human identity

The OWASP Non-Human Identities Top 10 describes NHIs as the identities software uses instead of people, such as service accounts, API clients and workloads, typically authenticating with API keys, tokens, encryption keys and certificates. Common types include:

Type	Example	Typical secret
Service account	Windows service, database application user	Password
Workload role	Cloud VM or container role	Short-lived token
API client	Integration with a SaaS platform	API key or OAuth client secret
Automation	CI/CD pipeline, runbook engine	SSH key, deploy token
Device or certificate	mTLS client, IoT device	X.509 certificate

How it works

An NHI authenticates the same way a person does, but without MFA or a human to notice something odd. NIST SP 800-53 Rev. 5 control IA-9 requires services to be uniquely identified and authenticated before communicating, and IA-5(7) forbids unencrypted static authenticators embedded in applications. Good practice is to give each workload its own identity, scope it to least privilege, keep its secrets in a vault or replace them with short-lived credentials, and log what it does.

Why it matters

The 2025 OWASP list names the main risks: improper offboarding, secret leakage, vulnerable third-party NHIs, insecure authentication, overprivileged NHIs, insecure cloud deployment configurations, long-lived secrets, poor environment isolation, NHI reuse, and humans using NHIs. AWS recommends that workloads use temporary credentials through IAM roles, including workloads outside AWS through options such as IAM Roles Anywhere. The OWASP Secrets Management Cheat Sheet recommends dynamic secrets with short leases where possible.

Non-human vs human identities

	Human identity	Non-human identity
Sign-in	Interactive, often with MFA	Programmatic, with a secret or token
Lifecycle	Joiner, mover, leaver processes	Often created ad hoc, rarely retired
Behaviour	Varies by day	Usually regular, so deviations stand out

Examples

An application account used by a reporting tool suddenly reads a whole salary table from a new workstation.
A deploy key committed to a repository years ago still has write access to production.
An engineer signs in interactively with a service account’s password, hiding who did what.

How SWOT supports it

The Central Vault in SWOT PAM holds SSH keys, cloud keys and API tokens with rotation, and SWOT Cloud PAM issues short-lived cloud credentials instead of long-lived keys. SWOT Radar resolves directory, local and cloud identities and shows which accounts can log in to each host. SWOT DAM scores every database identity, application accounts included, against its own 30-day baseline. FlowForge keeps automation credentials encrypted and records every execution. See DevOps pipelines.

Questions

Are AI agents non-human identities?: Yes, when they act with their own credentials. An AI agent that calls APIs, queries databases or runs commands needs an identity and secrets, and it raises the same questions as any other workload: what can it reach, how are its secrets stored and rotated, who owns it, and is its activity recorded and reviewed.
Who should own a non-human identity?: Every NHI should have a named human or team owner who is accountable for its purpose, permissions and lifecycle. Ownership makes reviews possible and stops orphaned accounts: OWASP lists improper offboarding as the first risk in its Non-Human Identities Top 10, for identities left active after the service or person behind them has gone.