AI is turning the network into a security control

Three announcements from the past week reveal where enterprise infrastructure is heading.

First, Hewlett Packard Enterprise raised its networking growth outlook as AI demand accelerates.

The company also announced a $1.2 billion order from Vultr for AMD Helios AI systems.

Second, Cisco and Omdia reported rapid adoption of AgenticOps:

• 51% of surveyed organizations already use agentic AI in production network operations
• 82% accept some network changes without prior human approval
• 99% require safeguards before trusting agents to act

Third, NVIDIA introduced an architecture for monitoring and controlling AI agents through independent software and hardware enforcement.

Together, these developments connect AI, networking, and cybersecurity.

AI infrastructure requires more bandwidth, lower latency, and faster operations.

Agentic AI also changes network behavior.

Agents generate traffic, access applications, use machine identities, call APIs, and make decisions across connected systems.

Some agents will also configure networks, reroute traffic, isolate devices, and respond to incidents.

The network is no longer limited to connectivity.

The network increasingly provides:

• Visibility into agent activity
• Identity and access enforcement
• Segmentation between agents and systems
• Detection of abnormal behavior
• Policy controls over data and applications
• Independent intervention when agents exceed authority

My takeawayAI-ready infrastructure must also be secure-by-design infrastructure.

Organizations should stop treating AI, networking, and cybersecurity as separate strategies.

The three now form one architecture.

Is your network ready to connect AI agents, observe their behavior, and stop them when necessary?