Netskope introduces Skylight Agent Action Control for pre-execution policies

Netskope introduced Netskope Skylight AI Security, positioning it as a unified way to discover, govern, and protect AI use across employees, self-hosted models, agents, and data, including pre-execution control for agent actions.

Research Overview

The blog cites findings from Netskope’s 2026 AI Risk and Readiness Report, including that AI is deployed in 73% of organizations while only 7% have governance in place for it.

It also references data showing that 94% of organizations report gaps in AI activity visibility and only 6% have complete visibility into their AI pipeline.

Key Findings

The post highlights that nearly 38% of organizations reported having more than 100 AI agents deployed, and 81.7% plan to deploy more agents in the next 12 months.

It states that 91% of organizations cannot stop a risky agent action before it executes, and it links this challenge to authority drift, where an agent’s permissions expand beyond what was originally approved.

Product Update

Netskope Skylight is described as a named suite within the existing Netskope One platform rather than a separate stack, using the Zero Trust Engine, NewEdge Network, a single client, and a single console.

The company says Skylight’s components apply policies from one place across AI use cases covering employees, self-hosted models, agents, and associated data flows.

Technical Breakdown

The blog describes Netskope Skylight Agent Action Control as a new component intended to classify an agent’s intent, assign a risk level based on what the agent touches, and set granular policy profiles by agent type.

It says the control is meant to allow, alert, or block an action before execution, rather than after, and that it is integrated into the same platform console used for other Skylight components.

Operational Impact

For discovery, Skylight is described as building one inventory by identifying users, apps, models, agents, and data flow across cloud, endpoints, and networks.

For governance, the blog attributes governance gaps to real-time policy coverage and describes controls including account-level app control, adversarial testing gating model launches, risk profiles governing agent capabilities, and classification to flag sensitive data before model training.

Leadership Perspective

The blog frames the broader Skylight approach as covering multiple stages of AI use through a unified control plane, with discovery, governance, and protection implemented from one place.

It also describes protection at the point of use, including blocking jailbreak attempts against self-hosted models, moderating prompts and responses, real-time coaching for risky app use, and pre-execution stopping of destructive agent actions.

Blog Signals brief: Netskope Skylight AI Security expands Netskope’s unified AI security suite by extending coverage to agent action control before execution and reiterating a single control-plane approach for discovery, governance, and protection. This is a fact-based summary of the vendor blog.