AI already runs inside your organization. It’s in the tools your teams open every day, the code your developers ship, and the agents starting to act without a person approving every step. According to the 2026 AI Risk and Readiness Report, AI is deployed in 73% of organizations, while only 7% have governance in place for it.
Over the past year we have been building and launching a whole suite of AI security components, delivered inside the Netskope One platform to address this challenge. As the suite has grown, it became appropriate to give it a unifying name in order to recognise the extent of the offering; giving security teams one place to see and control AI use across employees, self-hosted models, agents, and data, instead of stitching that view together from different tools.
Netskope Skylight provides a complete picture and a unified control plane to secure your entire AI landscape.
Built on the Netskope One platform
Skylight doesn’t ask you to stand up a new stack. The Netskope One platform stays exactly as it is: the Zero Trust Engine for dynamic risk context, the NewEdge Network for connectivity and performance, one client to run, and one console to manage. Skylight’s components plug into that same unified platform, so a policy set in one place applies everywhere AI shows up.
Agent Action Control: the coverage you need for the next chapter
As well as getting a new name, today the Skylight suite was extended to cover yet another risk vector.
Agents are the newest and fastest-moving part of the AI risk picture. We are seeing AI agents that can read files, open a terminal, and push changes on their own proliferate throughout engineering teams (and agents are fast getting into the hands of non-technical teams too). In fact, earlier this year nearly 38% of organizations said they already had more than 100 agents deployed, and 81.7% of organisations plan to deploy more agents in the next 12 months. The actions of these agents require serious controls; and while some rogue agents make the headlines, even supposedly intended actions may be undesirable. Agents at risk of ‘authority drift’ (the term used when an AI agent’s permission footprint expands beyond what was originally approved), or whose harness or instructions are too vague or permissive are a significant and growing risk within the enterprise.
Yet 91% of organizations cannot stop a risky agent action before it executes.
And so today we’re launching Netskope Skylight Agent Action Control, the newest component in our AI security suite, built to classify the intent behind what an agent is trying to do, assign a risk level based on what it touches, and set granular policy profiles by agent type to allow, alert or block an action before it executes rather than after.
We built Agent Action Control the same way we built everything else in Skylight: inside the platform you already run, not as a separate tool with its own console to manage. Agent Action Control is the proof of how fast our unified suite is moving to keep your AI strategy moving forward securely. To go deeper on how these new capabilities work, read today’s companion blog dedicated to the new product.
Skylight: The full view
Here’s how Netskope Skylight secures AI across every stage of use: with a unified approach to discovery, governance, and protection, all from one control plane.
- Discover: 94% of organizations report gaps in AI activity visibility, and only 6% have complete visibility into their AI pipeline. Skylight discovers every user, app, model, agent, and data flow, in the cloud, on the endpoint, and across the network, building one inventory instead of partial views.
- Govern: AI is deployed in 73% of organizations, but only 7% govern it with real-time policy, and 91% cannot stop a risky agent action before it executes. Skylight governs before risk becomes an incident: account-level control over every app, adversarial testing that gates a model’s launch, risk profiles that decide what each agent can do, and classification that flags sensitive data before it ever trains a model.
- Protect: Jailbreak attempts against untested models succeed in as few as five interactions and 42 seconds, and although 98% of organizations now use AI, just 8% enforce data policy consistently inside AI environments. Skylight protects at the point of use: blocking jailbreak attempts against self-hosted models, protecting sensitive data and moderating content in prompts and responses, coaching risky app use in real time, and stopping a destructive agent action before it executes.
What’s ahead
AI adoption is not slowing down, and neither are the ways it creates risk. Netskope Skylight AI Security is built to grow with that pace of adoption, inside the platform you already run.
If you’re thinking about AI security today, start with Netskope Skylight. Explore Netskope Skylight AI Security or talk to your account team about what a complete AI security strategy looks like for your organization.
