Okta Agent SSO GA
AI Agents as First-Class Identities

On August 24, Okta announced the general availability of Agent SSO, bringing the open Cross App Access protocol into the identity product used by over 20,000 customers. AI agents are now registered as first-class identities in Universal Directory and governed alongside human employees, with short-lived tokens replacing static API keys. Yet only 34% of organizations apply the same security controls to agents as to humans.

Why Agent identity governance is now urgent

When an agent books your travel, updates your ticket, or queries customer records, what identity does it use to reach enterprise systems? The most common answer so far: hardcoded API keys or broad OAuth grants — long-lived, over-scoped, and invisible in use. Okta's research found that only 34% of organizations always apply the same security controls to their digital labor force as to their human labor force; the rest run like anonymous network traffic, with no centralized visibility, administrative oversight, or lifecycle audit trails.

That is the starting point Agent SSO targets. Okta says Agent SSO brings the open Cross App Access (XAA) standard directly into the identity product used by more than 20,000 customers, letting organizations model AI agents as first-class identities governed by centralized policy — just as SSO centralized human access decisions at the identity provider layer, agent authorization now moves from individual applications up to the enterprise identity provider.

The core mechanism: XAA and first-class identities

XAA is an extension of OAuth, formally incorporated as the Enterprise-Managed Authorization extension for MCP. Initially led by Okta, it has been adopted by more than 25 early adopters including Anthropic, Zoom, and Slack. When an XAA-supported agent such as Anthropic's Claude connects to an enterprise application, Agent SSO registers it as a first-class identity in Universal Directory, visible alongside human employees; administrators can assign, monitor, and update its security policies just as they would for any human worker.

Okta positions the product around two questions: "Where are my agents?" — by registering every supported agent as a first-class identity, visibility is centralized across platforms; and "What can they connect to?" — identity-governed, short-lived tokens replace hardcoded credentials and broad authorizations, so agents can connect only to sanctioned applications, APIs, tools, and MCP servers under least-privilege policies. The Okta Integration Network (OIN) provides out-of-the-box integrations including Anthropic (Claude), Asana, Atlassian, Canva, Datadog, Figma, Glean, Linear, Notion, Slack, and Supabase.

Product boundary: what Agent SSO does and does not cover

Third-party analysis (quasa.io, August 26) cautions against reading Agent SSO as "governing every agent." Coverage depends on both ends supporting XAA: the requesting agent, the target app/API/MCP server, and the connection role must all be compatible; otherwise another integration, control, or legacy credential is still needed. Agent SSO is included at no extra cost for customers on core SSO plans, but it establishes an identity and authorization foundation, not full lifecycle supervision — what an agent does within scope after obtaining a token is not judged by Agent SSO.

That is where Okta for AI Agents, GA since April 2026, comes in: broader agent discovery, lifecycle governance, access reviews, telemetry, and runtime controls. Enterprises should treat Agent SSO as a starting point rather than an end state — first solve "who can connect," then "what they can do once connected."

Implications and action items for enterprises

References:
· Okta press release (2026-08-24): Okta brings first-class identity to AI agents with Agent SSO
· Okta research (34% figure): AI Agents at Work 2026
· Independent analysis (2026-08-26): Okta Agent SSO Went GA August 24, but Coverage Is Limited
· Okta for AI Agents GA (2026-04-28): Okta for AI Agents is Now Generally Available