The ratio is 90:1. That is the number of non-human identities (NHI) to human employees inside the average enterprise, according to recent Cloud Security Alliance data. In some organizations, it reaches 144:1. We are no longer managing a workforce of people; we are managing a digital ecosystem where machines outnumber us by two orders of magnitude. Yet, 70% of these AI agents are granted higher privileges than their human counterparts, and only 28% of organizations can trace an AI agent's action back to a responsible individual. This is not a governance gap; it is a governance vacuum. And into that vacuum, Okta has just stepped with its Agent SSO, a move that signals the true start of the AI identity war.
For years, the narrative in enterprise security has been about human identity. Zero Trust, multi-factor authentication, and single sign-on were all designed for the human sitting at a keyboard. But the architecture of the modern enterprise has shifted. AI agents are now writing code, answering customer tickets, and executing financial trades. They are not tools; they are digital employees. The problem is that they have been operating without IDs, without oversight, and without a lifecycle. Okta's announcement on August 2026 aims to change this by formally enrolling these agents into the corporate directory. This is not a feature update; it is a fundamental re-architecting of how we define the 'workforce.'
The Core: XAA and the Standardization of Machine Identity
The technical engine behind this shift is the eXtensible Agent Authorization (XAA) protocol. Let me be clear about what this is and is not. This is not a new cryptographic invention. It is a compositional innovation, a systematic application of existing OAuth 2.0 Token Exchange (RFC 8693) and JWT-based Client Authentication (RFC 7523) to the specific workload of AI agents. The brilliance here is not in the math; it is in the semantics. XAA formalizes the concept of 'agent can act on behalf of principal,' creating a consistent identity pattern for machine-to-machine interactions.
The true milestone, however, is the integration with the Model Context Protocol (MCP). Anthropic's MCP has become the de facto standard for AI agent tool interoperability. By having XAA adopted into MCP's Enterprise-Managed Authorization extension, Okta has effectively pushed identity down from the application layer to the communication layer. This means any tool call made via MCP can now carry verifiable identity context. This is the kind of systemic interdependence that matters. It is not just about authenticating the agent; it is about authenticating the intent and the authorization chain behind every single API call.
From a security engineering perspective, the shift to short-lived tokens is the correct direction. Stored credentials, like API keys, are the single largest vulnerability in NHI security. A leaked key is a permanent backdoor. By replacing them with short-lived tokens that auto-rotate, Okta limits the blast radius of a single compromise to minutes, not years. This aligns with the broader industry trend seen in Google's BeyondCorp and AWS IAM Roles Anywhere. The technical foundation is sound, but the real moat is not the protocol. Any competitor can copy XAA. The moat is Okta's Universal Directory and its 18,000+ enterprise customers. The integration depth with existing lifecycle management and cloud security posture is where the value lies.
The Contrarian Angle: The Hidden Cost of 'Open' Standards
Here is the unreported angle. The narrative of XAA being an 'open, vendor-neutral' standard is a marketing construct. While Okta has brought partners like Cloudflare, Slack, and WorkOS into the fold, the standard's evolution is still tethered to Okta's product roadmap. True neutrality would require formal adoption by bodies like the IETF or OIDF as an RFC. That has not happened. This is a land grab disguised as collaboration. By making Agent SSO free and bundling it into the core product, Okta is not being generous; it is executing a classic Open Core strategy. The free tier is the loss leader to establish the standard, while the paid tier—governing 'non-XAA' legacy agents—is where the monetization begins.
This creates a subtle but powerful lock-in. It is no longer just about migrating your human identity stack; now your AI agent access policies are also bound to Okta. The switching cost has just increased exponentially. Furthermore, the free tier is a data acquisition play. By managing agent identities, Okta gains visibility into the behavioral metadata of enterprise AI—call patterns, tool graphs, and frequency. This is the 'AI call relationship graph,' a dataset that will be the foundation for future security analytics and compliance auditing. The company that controls this data layer holds immense strategic value.
The Competitive Landscape: A Two-Horse Race
The competitive matrix is now clear. This is a two-horse race between Okta and Microsoft Entra ID. Microsoft has the ecosystem advantage with Azure OpenAI, Copilot Studio, and over 500 million active Entra users. For enterprises deeply embedded in the Azure stack, Entra Agent ID is the path of least resistance. Okta's counter is neutrality. For the multi-cloud, multi-SaaS enterprise that fears platform lock-in, Okta offers a standardized, independent choice. The alliance with Anthropic is the key tell here. Anthropic's support for XAA is a direct competitive reaction to Microsoft's deep binding with OpenAI. This is not just an identity standards war; it is the identity-layer projection of the AI model market share war. The battle is between the 'OpenAI/Microsoft stack' and the 'Anthropic/Open Standard stack.'
The Takeaway: Watch the Standard, Not the Product
Predictability is a myth; only volatility is real. The volatility here is in the standard-setting process. The success of Okta's Agent SSO hinges entirely on whether XAA becomes a true industry standard. If it is adopted by major agent frameworks like LangChain and LlamaIndex as the default identity module, Okta will secure a decade of dominance. If Microsoft's ecosystem steamrolls it, Okta's AI identity business becomes a niche product. History does not repeat, but it rhymes in binary. We saw this with BlackBerry and the iPhone. BlackBerry had the enterprise standard for mobile device management, but the platform shift to iOS and Android rendered it obsolete. The question is not whether Okta's technology is good; it is whether a single identity provider can survive a platform war. The next 18 months will determine if we are moving to a world of standardized digital labor or a world of platform-captive agents. The code is being written now, and the audit has just begun.
