Okta and 11 Industry Rivals Build a Shared Security Architecture for AI Agents

Quick Facts

• Okta and 11 other companies, including AWS, Google Cloud, Salesforce, and CrowdStrike, formed the Blueprint Alliance to create a shared architecture for securing AI agents.
• A large regulated asset manager scanned its environment and found approximately 13,000 agents, only about 1,000 of which it considered valid.
• 88% of organizations report suspected or confirmed AI agent security incidents, yet only 22% treat AI agents as independent, identity-bearing entities.

AI agent governance dominated Okta's Oktane event this week, as the identity company unveiled a cross-industry coalition and new runtime controls designed to stop unmanaged agents before they cause damage. The event, covered by theCUBE's Krista Case and Rebecca Knight, surfaced three core themes that point to where enterprise AI security is heading.

The scale of the problem is large. Okta's AI Agents at Work 2026 report found that 92% of executives say autonomous AI agents are already in widespread or moderate use inside their organizations. Non-human identities, including bots, service accounts, and AI agents, already outnumber human users by 100 to 1 in a typical enterprise. In cloud-native environments, that ratio reaches 144 to 1.

A Coalition Around a Reference Architecture

Okta's answer is the Blueprint Alliance, announced alongside AWS, CrowdStrike, Databricks, Docker, Google Cloud, Lovable, Proofpoint, Salesforce, ServiceNow, Wiz, and Zscaler. GE Appliances and World Central Kitchen serve as strategic advisors. The alliance is built around six principles: treating every agent as a first-class identity, scoping access to the task rather than granting standing access, keeping delegation traceable, monitoring runtime behavior continuously, enabling instant and reversible containment, and ensuring governance adapts at the speed AI moves.

The coalition expands a narrower blueprint Okta introduced in March, which asked organizations three questions: where are their agents, what can those agents connect to, and what can they do? The new framework adds a fourth question: how do you respond?

Charlotte Wylie, senior vice president and deputy chief security officer at Okta, described the reasoning behind a collective approach. "Rather than trying to solve that problem alone, there is a recognition that we have to galvanize as an industry to solve this problem for our customers at a speed that they are scrambling to run after," she said. Okta president and COO Eric Kelleher told analysts after the keynote to expect "a second cohort" of members "in the coming weeks."

Runtime Enforcement and Kill Switches

Okta's Agent Gateway sits between an agent and its tool calls, enforcing policy and logging interactions in real time. Ric Smith, president of products and technology at Okta, called this a significant shift. "That is a very different transition for Okta," he told theCUBE. Expanded kill-switch functionality can revoke tokens and terminate active sessions if an agent is compromised.

Analysts noted one gap: the Blueprint stops at containment. Stopping a misbehaving agent prevents further damage but does not reverse changes the agent has already made. Recovery remains an open requirement for the industry.

Cross-Vendor Coordination in Practice

Alliance members plan to publish reference integrations and joint interoperability findings on a regular basis. The goal is for a threat spotted by one vendor's runtime monitor to trigger action across every other vendor's control plane, using a shared format.

1Password's Credential Broker supplies secrets to agents without exposing them to the underlying model, according to Nancy Wang, the company's chief technology officer. ServiceNow maps agents to the business processes they support and is working with Okta on agent identities and session tokens, according to Bhakti Pitre, vice president of AI platform security product at ServiceNow.

The Discovery Problem Is Bigger Than Expected

One example from the event illustrated how far most enterprises are from knowing what agents they already have. A large regulated asset manager scanned its environment and found approximately 13,000 agents, only about 1,000 of which it considered valid. The scan also turned up more than 500 MCP servers against a handful that had been approved.

Okta's data reinforces how widespread this blind spot is. Only 10% of organizations report having a well-developed strategy in place to manage AI agents, even as 91% say they are already using them. Seventy percent of identity-related security incidents in Okta's dataset are linked to autonomous AI activity.

Customers at the event urged pragmatism. "We're treating AI like it's a brand-new thing. But really, you're still talking about identity data in another technology tool," one attendee said. "If you have a governance process and a technology intake program, do that, use that."