.”
Thus, the challenge for security teams is to understand not only what an AI app or agent is connecting with, but also to understand what data the AI tool can access and which identities or permissions it is using, he said.
This is a main reason why the discussions around AI security are so often turning into an identity security conversation—with a focus on the need for protecting non-human identities that are leveraged by agents, according to Ebley.
“The reason identity is so relevant is because AI is just a massive exacerbation of service account security that we’ve had for a long time,” he said. “You’re literally allowing an account, an application, to run on your behalf—and sometimes [it’s] masquerading as you, and then other times it basically has its own access.”
Major implications stem from the fact that while traditional identity security often focuses on the user or account, AI tools use those permissions in a different way, Ebley noted.
With traditional identity security, “the context is just the idea of a user or an account—not the application where things are being leveraged, not other applications that might sit alongside that,” he said.
Ebley said this can create something akin to an “octopus scenario”—where an AI application is connected to multiple different tools with varying levels of access.
Ultimately, that’s why having strong visibility is so essential for being able to effectively secure the usage of AI and agentic tools, he said.
If an organization is able to understand everything that an AI tool is connected to, “I can view the prompts that are coming into this. I can understand the data that it’s accessing and the level of permissions that are there, and how those identities are actually handled,” Ebley said. “You need to be able to understand all those things. Trying to solve an AI security problem, while having just one piece of that, is an impossibility. It’s Sisyphean, to say the least.”
Originally published on May 15, 2026 by CRN.