Executive summary
Securing enterprise AI agents requires an identity-first architecture built on open standards like the Blueprint Alliance reference architecture and Cross App Access. By integrating Okta’s identity controls across cloud, endpoint, and application stacks, enterprise IT teams can discover shadow AI, enforce least-privilege access, and deploy a centralized agent kill switch.
Every team wants to move faster with AI, but more often than not, this speed comes at the cost of control. Agents are acting on behalf of users, calling APIs, and reaching sensitive data, bypassing identity controls entirely. What started as a problem with a handful of agents is now compounding faster than ever.
No single vendor secures the entirety of an enterprise’s ecosystem alone. That is why the Blueprint Alliance was established: to bring together a coalition of industry leaders focused on building a practical framework for securing and governing agents at enterprise scale.
The Blueprint Alliance reference architecture for AI security
The Alliance brings together experts from across the technology ecosystem, including identity, data, applications, infrastructure, and cybersecurity, with each contributing their specialized knowledge. Together, the Alliance has coauthored a standardized reference architecture designed to help organizations answer four critical questions:
- Where are my agents?
- What can they do?
- What are they doing?
- How do I respond?
At Okta, our mission for the last 17 years has been to free everyone to safely use technology—a commitment trusted today by more than 20,000 customers. As agentic AI transforms the enterprise landscape, we believe identity remains foundational to answering these four critical questions. Quite simply, if you cannot securely identify and govern agents, you cannot protect the data they access or the actions they take. This is why Okta joined the Blueprint Alliance as a founding member.
We introduced Okta for AI Agents in 2025 to extend our identity platform to agents, allowing organizations to discover, govern, and protect them throughout their lifecycle. But while identity is the critical anchor, securing AI requires a connected stack. To put the blueprint reference architecture into practice, we have been building native integrations with our fellow founding Alliance members. These collaborations work together to secure your agents and help you answer those four defining questions.
Where are my agents?
Before you can secure agents, you need a complete inventory of what exists. Together with our partners, Okta discovers shadow AI and centralizes agent registration across a wide range of environments.
- Build and manage agents with Amazon Web Services (AWS): Okta and AWS have partnered on several integrations, enabling customers to build and scale AI agents on AWS while centrally governing their access policies in Okta. The integrations include:
- Okta MCP Server in AWS Marketplace: Manage your Okta environment directly from the conversational AI tools you already use every day.
- Amazon Bedrock AgentCore: Find, import, and register AI agents running on Amazon Bedrock, know what each can reach, and help enforce least-privilege access with a full audit trail.
- Detect shadow AI with CrowdStrike: Okta enables you to discover agents on endpoints using our integration with CrowdStrike, helping you detect shadow AI. When combined with Okta for AI Agents, organizations gain the ability to discover, onboard, protect, and govern any undiscovered AI use within their organization:
- Shadow AI Discovery for Endpoints: With the new CrowdStrike Falcon EDR integration, Okta provides visibility into AI agents running on endpoints, the systems they connect to, and their owners so you can bring them under your control.
- Find, import, and register agents from Databricks: As organizations continue to build AI agents within Databricks with Agent Bricks on the Databricks Data + AI Platform, Okta's AI Agent Registry allows them to be continuously imported and centralized for governance alongside the unified governance Unity Catalog and Unity Gateway provide across data, models, agents, and tools:
- Databricks Agents Agent Bricks: Find, import, and register AI agents built on Agent Bricks, the Databricks multi-harness agent development platform, know what each can reach, and enforce least-privilege access with a full audit trail.
- Find, import, and register agents from Google Cloud: Google Cloud provides advanced capabilities for long-running agents and agent reporting. Now organizations can register and govern those agents within Okta for AI Agents.
- Gemini Agent Platform: Find, import, and register AI agents built on the Gemini Agent Platform, know what each can reach, and enforce least-privilege access with a full audit trail.
What can my agents do, and what are they doing?
Discovering and governing agents is only part of the challenge. Organizations also need to centrally define what these agents can access, rather than relying on unmanaged service accounts, static API keys, or repeated user consent prompts.
Agent SSO, built on Cross App Access
An extension of OAuth, Cross App Access (XAA) is the open industry standard spearheaded by Okta. It’s recognized as enterprise-managed authorization within the Model Context Protocol (MCP) spec. It shifts access control away from individual applications and end users to the enterprise identity provider (IdP). Instead of negotiating connections app-by-app, XAA enables the IdP to centrally manage, broker, and audit authorization decisions across SaaS applications, AI agents, and non-human identities.
Agent SSO is how Okta brings that industry standard into the product our customers already use. Included in all Workforce Single Sign-On (SSO) plans, Agent SSO gives organizations:
- The ability to model agents as first-class identities: Any XAA-supported agent connecting to your applications—like Claude, Cursor, or VS Code—is treated as a first-class identity. Assign policies directly to the agent and audit its actions, rather than letting it run as anonymous app traffic.
- Instant connectivity through the Okta Integration Network (OIN): Immediately leverage prebuilt XAA integrations across leading AI agents, SaaS applications, and MCP servers through the OIN, connecting requesting agents like Claude to essential enterprise resources such as Slack, Asana, Atlassian, Zoom, and Figma without custom wiring.
- Reduced consent fatigue: IT preapproves agent access policies at the identity layer, protecting employees from endless, repetitive OAuth consent screens and avoiding accidental over-granting of permissions.
- Immediate value included in your SSO plan: Enterprises can secure known XAA-supported agent connections today using their existing Workforce SSO plan, lowering the barrier to securing agentic workflows.
- A seamless upgrade path to Okta for AI Agents: Agent SSO is the starting point for agentic security. Because your agents are already modeled as first-class identities under Agent SSO, you can scale to Okta for AI Agents to unlock shadow AI discovery, runtime enforcement, full lifecycle governance, human-in-the-loop approvals, custom auth servers, agent-to-agent connections, and a kill switch—all without re-registering agents or rebuilding connections.
How do I respond when an agent poses a risk?
Once you’ve defined what agents can access, you still need to monitor for risk and act fast when something goes wrong. Your IdP must not only monitor for identity-based risks, but also ingest signals from a variety of platforms. This collective intelligence provides a defense-in-depth strategy in which you can centrally enforce a kill switch at the point of access to reduce the risk of a breach.
A kill switch for agent deactivation
Deploy an AI agent kill switch with ServiceNow (limited release): ServiceNow, the control tower for AI governance and management, detects malicious agent activity and initiates containment. Okta for AI Agents serves as the identity platform for AI agents and, upon receiving a signal from ServiceNow, revokes the agent’s session tokens in real time, stopping the agent and reducing potential harm.
And it doesn’t stop there. Leading security vendors across identity, endpoint, network, data security, and more are collaborating on more ways to share agent activity signals from across your stack so that you can turn those signals into coordinated action.
Securing AI requires an open, connected ecosystem
These integrations underscore a simple point: agents don't stay inside one vendor's walls, so your security can't either. Answering the blueprint's four questions requires an identity layer that integrates with the endpoint, cloud, data, and network tools you already use.
Choose a vendor that works across all of the technologies in your stack, so you can see every agent, control what it can do, and shut it down when it matters.
Access the Blueprint Alliance’s open reference architecture and multi-vendor standards that are driving the future of AI governance.
Ready to put this into practice?
Watch Putting the Blueprint Alliance into Practice to see how top tech leaders are teaming up to tackle identity and trust in the agentic era. Featuring insights from CrowdStrike and Okta, this on-demand webinar delivers an actionable framework and roadmap for managing non-human identity sprawl, implementing Zero Trust principles, and defending against AI identity risk.