Zscaler Introduces Security Tools for Enterprise AI Agents

Zscaler announced a set of product capabilities intended to help enterprises secure and govern agentic AI systems.

New controls for AI agent traffic

In a company press release, Zscaler said the new capabilities include AI Broker with an Agent Registry, Endpoint AI Security, and AI Access Graph. The company positioned the tools as part of its broader zero trust approach to securing enterprise environments where AI agents may act on behalf of users, applications, or business processes.

Zscaler said AI Broker is designed to mediate communications involving AI agents, including interactions that use emerging agent communication patterns. According to Investing.com, the product is intended to support Model Context Protocol and agent-to-agent communications, often abbreviated as MCP and A2A. Zscaler’s press release said the related Agent Registry is meant to provide visibility into approved and active agents.

The company’s stated goal is to give enterprises a way to identify AI agents, apply access controls, and manage which agents are permitted to connect to systems or data. That framing reflects a growing security concern: autonomous or semi-autonomous AI tools may create new paths for data exposure or unauthorized actions if they are not inventoried and governed.

Endpoint and access visibility

Zscaler also announced Endpoint AI Security, which it described as a way to address AI-related risks at the device level. Investing.com reported that the product is intended to detect and manage device-level AI threats. Zscaler’s announcement said the capability is part of its effort to secure how AI tools are used from endpoints, where employees may interact with AI applications, agents, files, and enterprise data.

Another component, AI Access Graph, is intended to map access relationships involving users, applications, data, and AI agents. Zscaler said this graph-based view is designed to help organizations understand how access is being granted and where exposure may exist. The company’s release presents the feature as a governance tool for assessing AI-related access paths rather than as a standalone security product.

Network World reported that Zscaler unveiled the zero trust platform for agentic AI at Zenith Live 2026 and identified AI Broker, Endpoint AI Security, and AI Access Graph as central components of the launch.

Why enterprises are paying attention

The announcement comes as more vendors and enterprises experiment with AI agents that can retrieve information, interact with software, and perform multi-step tasks. These systems can be useful in business workflows, but they also complicate traditional security models because an agent may request access, call tools, or move data in ways that differ from normal human activity.

Zscaler’s announcement focuses on applying identity, access control, traffic mediation, endpoint monitoring, and visibility to that problem. The company did not present the products as eliminating all AI security risks. Instead, its materials describe a set of controls intended to help organizations manage agent behavior and reduce exposure as AI agents become more common in enterprise systems.

The core claim across Zscaler’s release and coverage from Network World and Investing.com is that enterprises need security infrastructure that can recognize and govern AI agents directly, rather than treating them only as ordinary applications or user sessions.