
In this exclusive article for DCNN, Jamie Moles (pictured above), Senior Technical Manager at ExtraHop, outlines how real-time network visibility, governed AI agents, and flexible model architectures can help data centres counter increasingly sophisticated cyber threats:
State-backed hackers from Russia, China, and Iran are shifting tactics against critical national infrastructure. Rather than striking immediately, they establish quiet, long-term access, planting dormant controls to trigger during future crises. These intrusions allow threats to cause devastating disruption in minutes, not days.
Whilst AI threats accelerate time-to-impact, expanding AI infrastructure creates new hiding places. Workloads are forcing data centres and neoclouds to upgrade from 100Gbps to 400Gbps networks. Legacy visibility tools cannot capture this volume, creating blind spots where dormant intrusions lurk undetected.
This dual pressure forces security leaders to re-evaluate their SOCs. Teams require real-time visibility into all AI activity to ensure legitimate workloads behave as intended and to spot anomalous behaviours. Batch-orientated SOC workflows are now too slow, but solving this speed issue requires more than plugging an LLM into legacy systems. Without rich context, you simply automate failure.
Defending against AI-powered threats requires a new architecture: real-time observation of all enterprise traffic, behavioural detections triggering governed actions, and pre-correlated context empowering agentic operations alongside human oversight. This is the foundation of the agentic SOC, where analysts shift from manual triage to directing agents that operate at machine speed.
AI agents are only as accurate as their context. Unreliable inputs (like alterable log data) produce wrong or inconclusive results. Agents need a structured context layer fusing real-time evidence with threat intelligence. Network telemetry forms this backbone. Because attackers cannot alter wire data, it serves as immutable ground truth. By structuring traffic into a pre-correlated context graph, detections across endpoints and identities are tied directly to actual network behaviour, letting agents and analysts act at threat speed.
Machine-speed defence does not require granting agents unchecked authority; recent incidents demonstrate that even advanced models make errors or find unintended shortcuts. Governance demands independent, live tracking of what agents access, move, and execute on the network. This oversight allows security teams to harness rapid response without surrendering control.
Because models continually evolve, relying on a single LLM is a flawed strategy. Different models excel at different tasks: some at detection, others at response. The underlying context and control framework form the true foundation, and the models are simply interchangeable components built on top.
Matching the speed of modern threats requires building the right foundation first. Open context, strong governance, and model flexibility make machine-speed defence both trustworthy and effective.
For more from ExtraHop, click here.

Head office & Accounts:
Suite 14, 6-8 Revenge Road, Lordswood
Kent ME5 8UD
T: +44 (0)1634 673163
F: +44 (0)1634 673173