What agent-driven AI means for network and security design

With the transition of agentic AI systems from trial pilot deployments into real-world settings, a significant limiting factor on the ability to deploy them in a production environment is the stability and security of the infrastructure. At Cisco Live EMEA 2026 in Amsterdam, Cisco has introduced several updates to infrastructure that are focused on the way in which large AI systems are networked, monitored, and secured when operating under various operational and regulatory constraints.

There have been three main areas of focus in regard to the announcements made:

(i)data center networking for AI workloads.

(ii) operational control of agentic systems.

(ii) Security mechanisms for autonomous AI interaction.

Networking design gets a reboot to stabilize AI workload chaos

Cisco has just launched the Silicon One G300, a new switch silicon designed to make high-density AI clusters work with ease whether you're training, inferring, or crunching data in real time. According to Cisco, one of the main problems here has been network congestion, which leaves jobs waiting & wasting resources.

Cisco reckons G300 manages to squeeze an extra 33% out of network use and cuts job completion times by a whopping 28% when compared to how things are normally laid out. And from an operational point of view, the idea here is to make AI job execution less unpredictable rather than just pushing the limits on peak performance.

G300 is the brain (or rather the silicon) behind the new Nexus 9100 and Cisco 8000 systems designed to deal with the big, tightly coupled AI jobs found in hyperscale data centers, on-site private clusters, and environments where things have to be super secure. Cisco has also just launched Nexus One, a management tool that's supposed to help simplify configuration and monitoring across on-prem and cloud-based data centers and that addresses all the issues around configuration drift and inconsistent operations.

Getting operational visibility in agent-driven systems

Cisco has also given us the lowdown on what's new with AgenticOps, an operational framework that's all about cross-domain telemetry and that's got its eyes on data from all sorts of places, like networking, security, and management. All this is coming in from places like Splunk and Nexus One.

The aim is to spot when things are going a bit haywire in environments where AI agents are making all the decisions. Traditional monitoring tools tend to be pretty reactive & focused on individual machines. But AgenticOps is all about correlating everything & seeing the big picture, and that's more and more important as AI systems start influencing infrastructure & external tools all the time.

This whole approach reflects the changing face of enterprise operations, where AI-driven environments need a constant stream of 'yes' or 'no' feedback rather than just fiddling with static settings.

Security controls are heading towards a new frontier with intent analysis at the wheel

Security updates are all about risks that appear with more and more "smart" AI systems, particularly the things people do with them and the way models can get messed with. Cisco is announcing some big changes to Cisco AI Defense; it's getting some new governance controls to more easily manage the supply chain side of things & also some extra protection to keep an eye on what these tool-driven systems are getting up to in real time.

Out on the network edge, Cisco is giving an upgrade to its Secure Access Service Edge (SASE) platform, making this new inspection system that actually looks at what an AI-generated request is hoping to achieve, not just what traffic looks like. This is a big goal to catch all the misuse that doesn't actually look like a conventional attack.

This all boils down to acknowledging that systems, which can think for themselves, are putting security at risk even if what they're trying to do looks perfectly normal.

Support models need to fit into places that have strict rules

Cisco has been talking about a lot of expanded Critical National Services Centers (CNSCs) across Europe in places in Germany, the UK, France, Spain, & Italy. Again, these centers have a whole separate way of working with tech support and approved communication channels to make sure data stays where it's meant to be and that there is operational isolation & complete transparency.

This model is being built specifically for places where data stays in the country, operations are isolated and auditable, & all are very important for the region.

In general, it's safe to say that Cisco’s updates are showing them to be sensible folk—that the biggest problems now aren't so much the AI tech they use, but making sure the underlying infrastructure is solid and operations are manageable, or else the whole thing falls over.

More For You

AI agent skills are quietly becoming a major security risk

Are Hackers Targeting Windows First While Macs Fly Under the Radar in India?

Your AI assistant has full access to your life and that should worry you

WhatsApp introduces a lockdown-style mode to reduce cyber risks