AI Agents Gain Write Access to AV-over-IP Systems

A staffer types a plain-English request into a chat window. A video wall on the demo stand reroutes itself, a multi-window layout rebuilds, and a log scrolls past showing what changed and why. No API documentation, no scripting. Until recently that was a concept video; in 2026 it is a shipping product. DVIGear's DisplayNet Connect, announced in May and demonstrated live at InfoComm 2026, runs a Model Context Protocol (MCP) server between an AI agent and the DisplayNet API, so assistants such as Claude and OpenAI Codex can design, deploy and diagnose an AV-over-IP signal distribution system in natural language. The pattern is spreading: AI is moving from a novelty in the meeting room to an operator with a seat at the control console. For integrators, the interesting question is no longer whether AI belongs in the signal chain, but what happens when the command is wrong.

The appeal is easy to see. AV-over-IP deployments have always demanded a scarce combination of networking and API skills — multicast design, IGMP snooping, VLAN segmentation, controller scripting and REST calls. An agent that can turn "route source 4 to displays 1 through 6 and rebuild the wall layout" into a valid configuration removes a genuine barrier to adoption, and it does the same for troubleshooting: reading logs, correlating events and proposing fixes are exactly the tasks where a tireless assistant adds value. Vendors frame the capability as design, deploy and diagnose, and for smaller integrators it can compress days of configuration into minutes.

It also changes the security model, because these are not read-only conveniences. Translating a request into a configuration and pushing it to a management system is a state-changing action — write access, in effect, granted to software that may be processing content it has no reliable way to trust. The MCP specification itself acknowledges that it cannot enforce security principles at the protocol level, and guidance from the NSA highlights prompt injection, weak approval workflows, insufficient access control and manipulated integrations as live risks. Indirect prompt injection is the sharpest example: malicious instructions planted in a document, a service ticket or a public repository can be read by an authorised agent and turned into commands against production infrastructure. In an AV system the worst case is not a lost file but a dark control room.

The practical response is governance rather than abstinence. Integrators should scope agent permissions to the smallest useful set, require human confirmation before any change to live routing, keep the control plane on a segmented management network, log every action and be able to roll back. It also helps to keep the media path architecturally simple. Amtune's audio-video decoder and AV-over-IP codec range transports video across standard IP infrastructure, the AV switch family handles routine routing, and the HD video seamless hybrid matrix provides a deterministic local switching layer. Because the signal chain is explicit and hardware-defined, an agent or a cloud dashboard can orchestrate it without owning the transport itself — precisely the separation of duties that AI-era AV security requires.