Decoding the world of cybersecurity

Okta expands AI agent identity across providers

Okta is extending enterprise AI identity controls across provider boundaries. New support for compatible external identity systems allows organisations to govern agent access without replacing established authentication platforms.

Okta expands AI agent identity across providers
Summary
  • Okta for AI Agents now supports compatible third-party identity providers using Cross App Access, including implementations from Ping, Descope and Oracle.
  • Organisations can manage agent permissions, access reviews and audit records for Okta-protected resources across connected identity systems.
  • The integration is generally available, although deployment requires compatible providers and does not replace application-level controls over an agent's behaviour.

Enterprises using several identity platforms can now govern access by AI agents through a common set of controls, following an expansion of Okta’s support for third-party authentication systems.

Okta confirmed on 13 November that third-party identity provider support is generally available in Okta for AI Agents and Agent SSO. The capability allows agents associated with compatible external identity systems to access protected applications through policies managed within the Okta environment.

The integration uses Cross App Access (XAA), a protocol intended to establish controlled connections between applications and automated agents. Compatible identity providers include implementations from Ping, Descope and Oracle, allowing customers to retain existing identity systems rather than immediately migrating every user and agent to one platform.

Access is limited to participating identity providers and supported resources. The release does not establish universal compatibility with every enterprise application or authentication platform.

Agents cross existing identity boundaries

Large organisations frequently maintain several identity systems as a result of acquisitions, regional operations or technology decisions made by individual business units.

Those arrangements can complicate employee access, particularly where applications belong to different subsidiaries or external partners. The introduction of AI agents adds another layer because an automated identity may need to perform tasks across systems managed by several organisations.

An agent operating on behalf of an employee may, for example, need information from a service protected by an identity provider different from the one used to authenticate that employee.

Without compatible authorisation arrangements, organisations may need custom connectors or additional service credentials to establish the connection. Each integration creates decisions about who owns the agent, how permissions are granted and which organisation can revoke access.

Okta’s new support allows compatible identity providers to establish trust using Cross App Access. An agent can retain its identity within the originating provider while obtaining controlled access to applications protected through Okta.

When an authorised employee initiates an agent task, the integration can validate the relevant identity and establish permissions for the protected resource. The resulting access can be scoped to the approved connection rather than granting the agent unrestricted authority across an organisation’s applications.

Employees can also continue using their established sign-in arrangements, reducing the need to repeat separate consent and authentication steps for each agent connection.

Shared governance extends to automated identities

Okta says agents from connected identity providers can appear within a shared directory, allowing administrators to identify their ownership and review the resources they can access.

The platform also provides consolidated audit records and access reviews, creating a common governance view across participating identity systems.

These capabilities address a growing difficulty as organisations deploy AI agents created by internal development teams, software vendors and business users. Automated identities may remain active after their original owner changes role or after the application for which they were created is no longer required.

Conventional identity governance processes depend on knowing which accounts exist, who is responsible for them and which permissions remain necessary. Applying similar controls to agents can make excess privileges and abandoned connections easier to identify.

Centralised records also support incident investigations. Where an agent accesses sensitive information or performs an unexpected operation, administrators need to establish which identity was involved, which permissions were available and the systems that permitted the connection.

However, a valid identity does not guarantee appropriate behaviour. An agent may operate within its authorised permissions while producing inaccurate results or performing a task that its owner did not intend.

Identity governance must therefore operate alongside application permissions, monitoring and controls over consequential actions. The new integration addresses how approved agents reach protected resources, rather than independently verifying every decision made after access is granted.

Acquisitions create a practical integration requirement

The commercial application is particularly relevant to organisations managing acquisitions and corporate restructuring.

Combining separate identity platforms can require substantial work involving user directories, authentication policies, application integrations and compliance controls. Those migrations may extend long after the acquisition itself is completed.

Where a newly acquired business already operates AI agents, replacing its identity provider before the agents can access central applications may delay integration and introduce additional operational risk.

Cross-provider support creates an alternative in which the existing identity system remains operational while approved agent connections are managed across organisational boundaries.

That approach still requires administrative work. Participating systems need compatible implementations, established trust relationships, appropriate user and agent mapping, and clearly defined permissions for the applications involved.

Incomplete onboarding can leave some agents outside the intended governance framework, while inconsistent access rules across providers can complicate investigations and periodic reviews.

Organisations must also determine which platform is authoritative for particular identity attributes and how changes are communicated when employees leave, agents are retired or permissions are revoked.

The new capability builds on Okta’s existing AI agent identity platform rather than replacing it. The development announced on 13 November is the general availability of third-party provider interoperability within that offering.

Okta’s product information confirms that connected agents can be brought into a shared directory and governed through scoped access, audit records and access reviews. Those controls apply to supported integrations; the company has not established that every identity provider or enterprise application can participate without further development.

For organisations operating several identity platforms, the release provides an additional way to manage agent access while retaining existing authentication investments. Its effectiveness will depend on the completeness of integration and the policies applied to each protected resource.

×