No accountable owner
A shared service account cannot tell you which agent acted, or which person is responsible for it.
HelixIAM AI Identity · sovereign by design
Give every AI agent its own identity with an accountable owner and a lifecycle. When an agent acts for a user, the token names both, and its authority only ever narrows. Workloads get identity without stored secrets.
Sovereign by design · your infrastructure in Europe · your keys · a Dutch vendor
POST /realms/acme/oauth2/token grant_type=urn:ietf:params:oauth:grant-type:token-exchange subject_token=eyJ… # alice actor_token=eyJ… # agent: invoice-assistant scope=invoices:read
{
"sub": "alice",
"act": { "sub": "agent:invoice-assistant" },
"scope": "invoices:read",
"aud": "https://api.acme.example/invoices",
"iss": "https://id.acme.example/realms/acme"
}actThe agent is named in every token; the user stays the subject.scopeAuthority narrows to what the agent needs. It never widens.ownerSuspend the agent and it gets no new tokens.The problem
Shared API keys and copied user tokens leave three questions without an answer: who did this, for whom, and how do we stop it without breaking everything else?
A shared service account cannot tell you which agent acted, or which person is responsible for it.
When an agent reuses a person's token, every action looks like the person did it — with all of their rights.
Rotating a shared key to stop one agent breaks every other system that uses it.
How it works
Each agent is its own identity in a realm, with an accountable owner, a purpose, roles of its own and an expiry date.
The agent exchanges the user's token for a new one through standard OAuth token exchange (RFC 8693): sub is the user, act is the agent.
Delegated authority is the intersection of what the user and the agent may do, restricted to the requested scope and audience. It never widens.
Agents whose owner has left are flagged for review. Suspend an agent and it gets no new tokens; resource servers that introspect see it at once.
Sovereign by design
Microsoft, Okta and Auth0 all sell AI-agent identity, and they do it well. They deliver it as cloud services operated by US-headquartered companies. HelixIAM gives you agent identity on your own infrastructure in Europe, with tokens signed by keys you hold.
| Criterion | Cloud agent-identity services | HelixIAM AI Identity |
|---|---|---|
| Products | Microsoft Entra Agent ID; Okta and Auth0 for AI Agents | HelixIAM, commercial software from KubeDNA B.V. |
| Where it runs | Cloud services operated by the vendor (Auth0 also lists a private deployment option on its Enterprise plan) | Self-hosted on your own infrastructure: you choose where it runs |
| Vendor | US-headquartered companies | KubeDNA B.V., a Dutch company |
Based on the vendors' public documentation, checked October 2026: Microsoft Entra Agent ID, Okta Agent SSO, Auth0 for AI Agents, Auth0 plans. Product names are trademarks of their owners.
Workload identity
Your pods and pipelines already carry a token signed by their platform. HelixIAM verifies it against the issuer you trust and returns a short-lived HelixIAM token bound to the roles of one application. Nothing to store, nothing to rotate.
POST /realms/acme/workload-identity/token grant_type=urn:ietf:params:oauth:grant-type:token-exchange subject_token=eyJ… # signed by the CI platform
{
"sub": "service-account-deploy-pipeline",
"realm_access": { "roles": ["deployer"] },
"exp": 1798732800,
"iss": "https://id.acme.example/realms/acme"
}issTrusted issuer, subject and audience must all match, or the request is refused.expShort-lived, bound to the roles of one application.secretNothing to store or rotate in the pipeline.Governance
Owners, status, bound applications and last use in one console screen, next to the users, applications and keys of the same realm. Every agent token issued or refused — direct or delegated on behalf of a user — lands in the searchable audit log, together with lifecycle changes, admin actions and workload token issuance, and you can stream the log to your SIEM over HTTPS.
Built on standards
Ready for MCP: OAuth protected-resource metadata (RFC 9728) tells AI tool servers' clients where to get tokens.
Delegation uses RFC 8693, so any OAuth-aware resource server can read the act claim.
Register clients through the standard API, with admin-controlled policy.
Deployment pattern
You don't need to replace your workforce IdP. People keep signing in where they do today and are brokered into HelixIAM over OpenID Connect or SAML; HelixIAM governs the agents and workloads that act for them, on infrastructure and keys you control.
sub and the agent as act.Early access/now onboarding launch partners
A paid pilot on your own Kubernetes: a few real agents or MCP tool servers registered with owners, one delegated user journey, keyless identity for one pipeline, a suspension drill and audit export to your SIEM. Then a production plan.