No Single Vendor Is Going to Be the AI Control Plane
September 23, 2026 · Keith Pasley
The AI control plane is starting to look less like a product and more like an architecture.
That is what caught my attention about the new Blueprint Alliance announced on Tuesday September 22 2026.
AWS, Google Cloud, Okta, CrowdStrike, Databricks, Salesforce, ServiceNow, Proofpoint, Wiz, Zscaler and others are working on a shared architecture for securing AI agents across the enterprise.
That matters because an enterprise agent rarely lives inside one technology stack.
Its identity might come from one platform. Its model from another. Its data could sit across several clouds. It may invoke SaaS applications, APIs, databases and other agents. Security telemetry could come from several more systems.
So who actually controls the agent?
I don’t think the answer will be one vendor.
The architecture needs to connect identity, authorization, data governance, runtime security, observability and response.
This is also why the distinction between AI governance and AI security is getting harder to maintain.
Governance can define what an agent should be allowed to do.
Identity can establish who or what the agent is acting as.
Runtime security can determine what it is actually doing.
Observability can provide the evidence.
Enforcement can stop the action when something crosses the boundary.
Those controls need to exchange context.
The Blueprint Alliance is proposing four questions that I think are useful for almost any enterprise agent architecture:
Where are my agents?
What can they do?
What are they doing?
How do I respond?
The technology underneath those answers will probably come from multiple vendors.
The architectural challenge is making those controls operate as one system.
Why this matters: Agentic AI crosses identity, data, applications, APIs, models and infrastructure. Security controls that remain isolated will have an incomplete view of the agent.
Decision it changes: I would be cautious about choosing an “AI control plane” based solely on which vendor has the longest feature list. Interoperability and the ability to exchange policy, identity, telemetry and enforcement signals may matter more.
Who should care: CIOs, CISOs, enterprise architects, AI platform leaders and security architects.
What I would ask next: If your AI agent crosses five technology platforms to complete one business task, which system has enough context to decide whether that action should be allowed?