Agentic AI blows up the attack surface as security moves into the infrastructure layer
SiliconANGLE Ryan Stevens ● Covered by 3 sources
Agentic AI is making enterprise networks harder to defend. Broadcom says security has to move into the infrastructure, not sit at the perimeter.
Based on reporting by SiliconANGLE, Ryan Stevens — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Agentic AI is pushing enterprise security down into the guts of the cloud. That’s the basic message from Broadcom’s Umesh Mahajan, who argues that perimeter tools can’t keep up with fast-changing AI workloads and the traffic they create between systems.
He’s not talking about a tweak. He’s talking about moving enforcement into the virtualization layer, so policy can be applied without forcing traffic through extra security boxes. Mahajan said the old “I’ve got a firewall, I’m fine” mindset no longer works, because attackers can get around it. His view is blunt: lateral security can’t be treated as a long project anymore.
The bigger issue is that enterprises have spent years buying security products one by one, then leaving the seams exposed. Mahajan compared that to Swiss cheese: lots of pieces, lots of holes. Broadcom’s pitch is an integrated stack built around vDefend and Avi Load Balancer, with the parts sharing context instead of operating as disconnected tools.
Scale is what makes this urgent. AI workloads generate heavy east-west traffic, and any inspection layer that adds too much delay becomes a problem fast. Mahajan said Broadcom is handling 75 terabits per vCenter cluster for firewalling and 17 terabits for IDS IPS, while keeping latency low by doing the work at the hypervisor level.
There’s also a visibility problem. Agents and Model Context Protocol services can appear and disappear quickly, so administrators need a real-time view of what’s approved and what isn’t before they can quarantine anything. Mahajan’s warning was simple: if security waits until production AI traffic is already flowing, the organization is already behind.
My take — AI-written commentary, not fact-checked reporting
This is what happens when cloud security gets treated like a shopping list instead of a system. The industry keeps bolting on tools and then acts surprised when the gaps become the attack surface. Broadcom is right about one thing: if AI traffic is moving this fast, security can’t live in a polite little perimeter anymore.
Read more about this at: SiliconANGLE