Agentic AI absorbs how a business really works, which makes that learned knowledge a new perimeter. Dr. Sean Murphy of F5 explains the risks and the platform approach to protecting it.
Security keeps declaring new perimeters: the network edge, data in the cloud, the API, workforce identity. Agentic AI adds one that is not a technical boundary at all. It is what your organization knows.
Dr. Sean Murphy of F5 argues that autonomous agents, once connected to internal systems, data lakes and operational workflows, build a detailed picture of what makes a business successful. He calls that learned knowledge your "alpha." The risk is that it becomes your omega if the agent is compromised.
The piece starts with why this differs from protecting customer lists or product designs. A breached agent could expose the logic and strategy behind the business, not just a file. It then turns to the threats: prompt injection, model poisoning and autonomous escalation. An attacker who manipulates a live agent could quietly rewrite business rules or pull proprietary training datasets from internal databases, bypassing controls built for data at rest or in transit.
From there, it sets out a response built around applications and APIs. That means web application and API protection, a web application firewall with AI built in, and proactive AI red teaming to find weaknesses before attackers do. It also covers a unified control plane, which matters to a CISO who would otherwise manage a dozen separate tools and the gaps between them.
For security leaders, the takeaway is a shift in what to protect: the dynamic processes and interactions of AI agents, not only static data.
This is an Experts Corner on ITSPmagazine. Learn and view more stories at https://www.itspmagazine.com/blog?topic=experts-corner
If you want to contribute your own story to the Experts Corner, visit https://experts.itspmagazine.com/