AI Agent Security: Risks You Cannot Ignore

Admiral Mike Rogers on AI Agents and the Next Cybersecurity Challenge

AI agents promise speed, efficiency, and scale. But as organizations race to deploy them, many are underestimating the security implications that come with autonomous systems operating across environments and data sets.

In the second part of Episode 1 of Identity Security Unleashed for the AI Era, Akeyless CEO Oded Hareven sat down with Admiral Mike Rogers, former Director of the NSA and Commander of US Cyber Command, for a strategic conversation on the emerging cybersecurity risks tied to AI agents. Drawing on decades of experience defending and penetrating complex systems, Rogers offered a sober perspective on where security teams should focus next.

Besides identity, the things that you’re hearing more with regards to other concerns with AI agents, what other aspects, do you hear about? Everybody’s trying to maximize the effectiveness of AI and how they’re applying it. And so clearly, we’re we’re the next kind of evolution you’re seeing is this idea of AI agents as vehicles to create massive speed and scale for us, to make us much more efficient, much more effective, if you will, in how we’re applying artificial intelligence within our organizations. But the challenge with that to me is so you look at those AI agents, look it, you’re talking about autonomy, you’re talking about reason and reasoning and planning, you’re talking adaptability. Those are all very good things. Don’t get me wrong. We’re gonna build much of our operational concept around maximizing the application of these agents, then protecting their security and protecting their identity becomes a really new challenge for us. I got really concerned about identity because my concern was guys, look, as an attacker, if I put on my other hat, compromising identity is one of the key techniques that we will often use. And we tend to focus on identity from a human perspective. Identity from a machine perspective is just exploding. And so identity from a machine perspective is becoming a massive problem.

Admiral Mike Rogers on AI agents and cybersecurity risk

Why AI Agents Raise the Stakes for Security Leaders

For Admiral Mike Rogers, AI agents do not introduce a new category of risk. They intensify the ones security teams already struggle with, especially complexity and identity.

“One of my biggest frustrations was complexity,” Rogers said, reflecting on his time leading both NSA and US Cyber Command. “The threat keeps getting more complex, but so do the solutions we’re deploying. If we’re going to get to speed and scale, we have to simplify, not make things harder.”

Identity sits at the center of that tension.  “If I put on my attacker hat, compromising identity is one of the key techniques we will often use,” Rogers said.

From the attacker’s perspective, identity is not limited to people. Yet many organizations still scope identity programs around human access first. That assumption no longer holds. Machine identities are growing faster than human ones, and AI agents accelerate that shift. “Identity from a machine perspective is becoming a massive problem,” Rogers warned, “and we need solutions that are simpler, not harder.”

Oded Hareven tied that reality directly to how AI agents operate. Agents are autonomous, adaptive, and deeply interconnected. They do not just respond to requests. They act. As Rogers noted, organizations are now building core operational models around these agents, which makes “protecting their security and protecting their identity a really new challenge.”

That challenge extends beyond access to data itself. Rogers described data as the fuel that powers AI agents. As agents become more capable, they require access to more information from more sources. “When you build high-performance engines,” he said, “your fuel consumption goes way up.” The same dynamic applies to AI. More autonomy demands more data, and that increases the consequences when identity controls fail.

Despite the gravity of these risks, Rogers was not pessimistic. He emphasized that the problem is solvable if organizations adjust their mindset. Security cannot focus solely on output and efficiency. They must design for risk from the beginning, not attempt to contain it later.

Key Takeaways

• Complexity is one of the biggest threats to effective security
• Identity compromise remains one of the most reliable attack techniques
• Machine and AI agent identities are growing faster than human identities
• AI agents increase both the amount of data accessed and the impact of failure
• Security must be designed into AI systems, not added after deployment

Security Before Speed

AI agents reward speed, but they punish mistakes at scale. As Admiral Rogers emphasized, autonomy and access demand discipline early.

In addition to this conversation, Episode 1 includes an address from Akeyless CEO Oded Hareven on why AI agents demand a new approach to identity, and a hands-on demonstration of secure, secretless AI agent access.

Watch Episode 1 to see it all.