Aembit Brings Okta Identity Policies Into Agent-Driven SaaS Connections
Identity security company Aembit has added support for Okta Cross App Access (XAA), giving enterprises a way to apply existing workforce identity policies when AI agents connect to SaaS applications and Model Context Protocol (MCP) resources.
The integration became generally available on September 22 alongside Oktane 2026, with Aembit participating as an Okta XAA launch partner.
The move addresses an emerging enterprise security problem: AI agents increasingly operate across multiple applications on behalf of employees, creating new authorization relationships that traditional user-centric identity systems were not designed to manage at scale.
XAA Moves Authorization Away From Repeated User Consent
Traditional OAuth workflows can require employees to approve individual application connections. That model becomes increasingly cumbersome when an AI agent needs to interact with numerous enterprise services during a single workflow.
Okta's XAA architecture moves more of that decision-making into centrally administered identity policy.
Administrators establish managed relationships between requesting applications and downstream resources, allowing approved agents to obtain access without repeatedly asking employees to authorize every connection.
The user's corporate identity remains central to the authorization decision.
Human and Agent Identities Remain Linked
Aembit adds another layer by combining the identity of the employee with the verified identity of the AI agent acting on that person's behalf.
This “blended identity” approach is designed to answer three critical questions: which agent accessed a resource, which user it represented, and which security policy permitted the action.
That distinction becomes increasingly important as autonomous systems move from simply generating information to retrieving corporate data and executing actions across business applications.
MCP Access Gains a Common Enforcement Layer
Aembit is implementing XAA through its new Enterprise-Managed Authorization Credential Provider, based on the MCP Working Group's Enterprise-Managed Authorization extension.
The provider allows Aembit to function as a policy enforcement point between agents and MCP servers, tools and other enterprise resources.
Organizations do not need every application in their environment to support XAA immediately.
Where XAA is available, Aembit can use the new authorization model. Where it is not, the platform can continue using conventional OAuth authorization, allowing companies to introduce the technology gradually rather than migrate their entire identity architecture at once.
Agentic AI Expands the Identity Attack Surface
The broader security significance lies in the rapid growth of non-human identities.
An AI agent may access email, development environments, project-management platforms, databases and other sensitive resources during one automated task. Poorly governed permissions could therefore give a compromised or misdirected agent access far beyond its intended role.
Centralizing authorization can help security teams enforce least privilege, revoke access and maintain audit trails without relying on individual employees to manage every agent-to-application connection.
Aembit's integration reflects a wider shift in enterprise identity security: organizations increasingly need to govern not only who a user is, but which autonomous software is acting for that user, what it can reach and what it is permitted to do once connected.