vietnamese crab exporter
Commercial Content

NordLayer launches NordLayer AI Scout to track what AI agents can actually access

Karen Koehler

NordLayer’s new NordLayer AI Scout aims to show security teams exactly what their coding agents can reach, including every MCP server they connect to.

That is an uncommon place to start a security product launch, but it can get at a problem enterprises are beginning to inherit as AI agents spread inside development teams. Approving an agent tells IT what software entered the company. It does not necessarily tell them what that agent can reach later.

NordLayer AI Scout, available now, is NordLayer’s attempt to make that activity visible. The platform tracks AI use across web and desktop, but its more technical features are aimed at the connections forming behind agentic tools. It keeps a live catalog of Model Context Protocol servers that coding agents actually connect to and inventories the skills available to those agents.

For security teams, that can create a different kind of asset map. An application may be approved once, while its connections keep changing as developers add tools or give agents access to new resources.

“We built NordLayer AI Scout for how AI actually works — not just what it looks like from the outside, but for what’s happening at a technical level,” said Vakaris Noreika, VP of product research and development at NordLayer. “Its AI DLP is designed to inspect prompts and control them pre-submission, to ensure that sensitive data is never inputted in the first place. Alongside it, the solution’s agent visibility maps what AI agents can access — resources, knowledge, and critical information — through a live MCP server catalog and a skills inventory for every AI agent in the organization.”

View the NordLayer demonstration here.

The agent is only the front door

Enterprise software inventories have traditionally been good at answering fairly static questions: what application is installed, who has access, which license was assigned.

Agents complicate that model because their useful functions often depend on what sits behind them.

A coding agent connected through MCP may gain access to repositories, internal knowledge, or other tools needed to complete tasks. Add another server and the agent’s reach changes. Give it another skill and its practical capabilities change again.

That makes the connection layer part of the security problem.

NordLayer AI Scout is designed to record those connections as they exist in real use rather than asking administrators to reconstruct them from provisioning records. NordLayer says the solution can show the resources and knowledge available to AI agents, giving security teams a clearer way to see where access has accumulated.

The concern is less dramatic than an agent suddenly running wild. More often, permissions grow gradually because each addition is useful at the time. Months later, the organization may have a much harder time explaining what a given agent can see.

Prompts create their own access path

NordLayer aims to apply the same logic to the information employees put into AI tools.

Instead of treating a prompt as ordinary data in transit, NordLayer AI Scout’s data loss prevention system can inspect it before submission. A company can configure the system to warn the user, redact sensitive information, or block the prompt before the material reaches the AI service.

That matters because employees do not need to upload a document to expose confidential information. A few copied lines of code or internal text can create the same problem.

Traditional DLP products were built around more familiar ways information moved outside an organization. AI prompts compress the action into a chat window, and the employee may not think of what they are doing as sending data anywhere at all.

Governance gets more technical

NordLayer AI Scout also monitors sanctioned and unauthorized AI tools across web and desktop, but NordLayer is framing visibility as more than a list of applications.

“What companies are facing today is an AI zoo — dozens of unsanctioned LLMs in daily use across the workforce, each one widening the organization’s exposure to data leaks and security breaches,” said Andrius Buinovskis, VP of product strategy at NordLayer. “True AI governance requires extensive visibility.”

The company’s argument is that governance increasingly depends on telemetry taken from real activity: which model was opened, what entered the prompt, which agent connected to which server, and what that agent was able to do once connected.

That is a more demanding job than maintaining an approved-software list.

NordLayer launched its enterprise browser in March, and NordLayer AI Scout pushes further into the controls surrounding AI use itself. For enterprises experimenting with agents, the timing is useful. The hardest question may soon be less “Which AI tools are approved?” and more “What have those tools been connected to since then?”