Non-human identities (NHI)
AI agents and non-human identities in Nova IAM
Nova IAM is an enterprise IGA platform for your entire IT landscape. Among non-human identities (NHI), it has managed AI agents as an identity type of their own since v0.9.4, alongside people and technical identities.
As of November 2026
Nova governs who is responsible for an agent, which capability roles it holds in Nova, who reviews that and when it ends.
In the product
What Nova can do today
The same tools apply to AI agents as to people: requests, approvals, recertification and logging. On top of that come rules that apply only to agents.
-
Owner and purpose
When an administrator creates an AI agent, an owner, who must be an existing identity of the type natural person, and a purpose are mandatory. Agents taken over through the Keycloak or Entra import start without an owner and a purpose, and the healthcheck reports missing owners. A review date and an expiry date are optional. Once the expiry date has passed, Nova runs the leaver routine, which by default sets the identity to inactive and locks its accounts.
-
Requests and approval
Only the agent’s owner, the owner’s deputy or an administrator can file a request for an AI agent. An approval workflow can name the “Agent Owner” as approver.
-
Recertification
A campaign with the scope “AI Agents” reviews the entitlements of the agents, and “Agent owner (AI identity)” can be chosen as reviewer. For each linked routine assistant, the matrix of its approved objects and actions is up for review as well. If it is revoked, Nova switches the routine assistant off.
-
Handover when the owner leaves
The default leaver workflow transfers the agents to the owner’s deputy, otherwise to their manager, and then to their line manager according to the organization. If Nova finds no successor, the agent stays without an owner and Nova writes a warning to the audit log. Nova logs every handover as well.
-
Service account in Keycloak
In Keycloak, Nova creates a confidential client with a service account for an agent, which signs in with client credentials. Tokens are valid for five minutes by default. “Rotate Secret” generates a new client secret. Nova shows it once and does not store it.
-
Missing owners and overdue reviews
The healthcheck reports technical identities and agents without an owner at severity high. The “Agent Review Scan” job writes overdue reviews to the audit log. An administrator sets up this job, and it does not send notifications.
Technical accounts and AI agents do not count as identities when it comes to size, see the pricing FAQ.
Limits
What Nova does not do today
Nova provides governance for AI agents, not runtime protection. Coverage of other non-human identities is limited as well.
Capability roles only take effect in Nova
Nova does not transfer roles with the nova-ai: prefix to target systems. They govern which objects and actions a linked routine assistant may use when assembling steps for lifecycle routines. The agent itself does not act in Nova.
No other entitlements through requests
When assigning and requesting, Nova only allows capability roles for AI agents. It rejects business roles and other entitlements. An agent’s rights in target systems therefore do not go through requests in Nova today. An import from Keycloak or Entra ID does, however, take over an agent’s groups as assignments.
Service accounts only in Keycloak
Nova only creates an agent as a service account in Keycloak. In other target systems, an administrator can only create an ordinary account for an agent.
No automatic discovery
Nova does not yet find service accounts and app registrations on its own and does not build an inventory of non-human identities from the target systems.
No runtime gateway and no prompt inspection
Nova does not sit between an agent and the systems it calls. It does not inspect prompts or filter what an agent does at runtime.
Mapping
OWASP NHI Top 10 (2025): where Nova stands
The OWASP project names ten risks for non-human identities. Nova partly covers three of them and does not cover seven.
| Risk | Status | In Nova today |
|---|---|---|
| NHI1 Improper Offboarding | Partly covered | Expiry date, owner handover when the owner leaves, and recertification with the “AI Agents” scope. Handover and scope apply to AI agents only. |
| NHI2 Secret Leakage | Not covered | Nova does not search for secrets in code, logs or configuration. |
| NHI3 Vulnerable Third-Party NHI | Not covered | Nova does not record third-party integrations and extensions or their access. |
| NHI4 Insecure Authentication | Partly covered | In Keycloak, the agent’s client signs in with client credentials. Nova does not check the authentication methods of other non-human identities. |
| NHI5 Overprivileged NHI | Not covered | Capability roles only limit which objects and actions the AI may use for a linked routine assistant. This does not cover excessive rights in target systems. |
| NHI6 Insecure Cloud Deployment Configurations | Not covered | Nova does not check the configuration of cloud or CI/CD environments. |
| NHI7 Long-Lived Secrets | Partly covered | In Keycloak, tokens are valid for five minutes by default and the client secret can be rotated. Someone has to trigger the rotation, as there is no schedule. Outside Keycloak, Nova does not cover this. |
| NHI8 Environment Isolation | Not covered | Nova does not check whether the same non-human identity is used across development, test and production. |
| NHI9 NHI Reuse | Not covered | Nova does not detect whether several applications or workloads share a non-human identity. |
| NHI10 Human Use of NHI | Not covered | Nova does not detect whether people use a non-human identity for manual work. |
Our own assessment, as of November 2026. OWASP has not reviewed it. Source of the list: OWASP Non-Human Identities Top 10.
Next step
Ask your questions live on the system.
In a 30-minute video call we show you on the demo system how Nova manages AI agents.
Request a demoRead on
- AI in Nova IAM
Operating modes, features and which data goes to the model.
- Roadmap
What is available and what is planned.
- Trust Center
Policies, mappings and the current audit status.