Agentic AI at the edge: The security frontier worth owning

A trend has emerged with quiet velocity and enormous consequence. Agentic AI, systems that do not merely assist but reason, plan, and act, has moved beyond the data centre and into the physical world. The cognitive layer that once powered enterprise chatbots is now being embedded into autonomous physical systems operating at the edge: robotics, warehousing, logistics networks, and critical infrastructure, in real time, without a human in the loop.

This is not a future scenario. The infrastructure to deploy it, open source, enterprise grade, air gapped, and hardware agnostic, is already available. Platforms like NemoClaw provide autonomous AI agent deployment with built-in policy engines, privacy routing, and network guardrails. The capability has arrived faster than the governance frameworks, faster than the security architecture, and considerably faster than most boards have been briefed.

The security perimeter did not just expand. It moved into the physical world.

For security practitioners, this represents a fundamentally new problem. In a traditional system you secure the hardware, the network, and the software stack. In an agentic physical AI system, the model itself, its reasoning, its planning loop, its tool use, is the system. The agent can perceive its environment, decompose complex problems, spin off sub-agents, execute tasks, and communicate externally. Compromise the agent, and you compromise the machine. In a warehouse, a logistics network, or a piece of critical infrastructure, that is not a data breach. It is an operational failure with physical consequences.

For executives, the challenge is equally urgent and considerably less understood. Agentic systems are entering enterprise environments through multiple channels simultaneously: open source deployments, hyperscaler platforms, and vendor-embedded AI in operational technology. The question is not whether agentic AI will run at the edge of your organisation. It already does. The question is whether your security architecture was designed with that in mind, or retrofitted onto infrastructure built for a different era.

This session examines the emerging security frontier that physical AI creates, drawing on direct experience securing AI and critical infrastructure systems across Asia Pacific and North America. It moves from the strategic to the practical: what the agentic architecture actually looks like, where the vulnerabilities sit, what a security framework designed for autonomous physical systems requires, and how organisations can move from reactive governance to genuine architectural advantage.

The organisations that will own this frontier are not the ones that treat physical AI as a risk to be managed. They are the ones that treat it as a design problem to be solved, early, deliberately, and with the right architectural foundations in place.

Agentic AI at the edge is not coming. It is here. The security community has the expertise, the instinct, and the frameworks to lead this conversation. This session is about making sure it does.