Agent Action Control is part of the Netskope Skylight portfolio, the new name for the former Netskope One AI Security. The family includes AI Command Center, AI Guardrails, AI Gateway, Agentic Broker, and AI Red Teaming.
AI agent actions
before they are executed and categorizes them into nine categories based on their purpose. These categories include access control changes, configuration modifications, credential and secret management, data destruction, infrastructure provisioning, potential information exfiltration, external communications, remote code execution, and source code modification.
Based on this classification, administrators can establish different policies for each type of agent and assign risk levels of low, medium, high, or critical.
Policies allow them to block or authorize a specific action, as well as generate an alert. When an operation is blocked, the system can send a notification to the user indicating that the action was not permitted.
Profiles can be configured individually for different types of agents. For example, a programming assistant can have different permissions than an AI-powered chat application.
Logging and Analysis of Operations
: All analyzed actions are logged. Security teams can filter alerts based on various parameters, such as cost exposure, source code modifications, infrastructure changes, or communications with external systems.
The technology operates on network traffic inspected by the Netskope platform, thus utilizing existing infrastructure and the management console.
One potential application occurs when a scheduling agent attempts to delete a production repository. In this scenario, the system can classify the operation as a critical data destruction and block the request before it reaches the repository.
Policies for Autonomous Agents:
The system is geared toward situations where an AI agent has permission to interact directly with enterprise systems. Policies allow you to predefine which operations each agent can perform and which need to be blocked.
This approach also aims to limit situations of so-called "authority drift," where an agent's effective permissions or capabilities can be expanded or used in ways not initially intended.
According to Netskope's 2026 AI Risk and Readiness Report, 91% of the organizations analyzed cannot stop a risky agent action before it is executed. Another study, Gravitee's The State of AI Agent Security 2026, indicates that 54% of the organizations surveyed reported a confirmed or suspected AI-related security incident in the past year.
