Concepts · 01 of 08
What is shadow AI, and how do security teams find it?
In short
Shadow AI is any use of AI tools, AI features or AI agents that the security team has not approved or cannot see. It is usually well-meant: people use what makes them faster. Teams find it by inspecting traffic, by inventorying what runs on devices and inside SaaS apps, and by asking.
By The Inline Desk · GenAI Security Finder · Published 2026-10-12
Desk research from the six vendors' own public pages, read on 7 October 2026. Profiles follow our published method; we did not run the products.
Where does shadow AI come from?
Most of it starts with someone trying to do their job faster. A marketer pastes a draft into a chat assistant, a developer installs a coding agent, a finance analyst turns on the AI feature that arrived in a spreadsheet tool last month. None of these needs a purchase order, and many arrive inside software the company already pays for. That is why shadow AI grows faster than shadow IT did: the tools are free or bundled, they run in a browser tab, and the benefit is immediate.
Vendors publish their own figures on how common it is. Netskope, quoting its Threat Labs report on generative AI, states that 72% of generative AI use in the enterprise is shadow AI. Treat figures like this as one vendor's research, not a benchmark for your company.
Source: Netskope: securing generative AI · Read 7 Oct 2026
Why is it a security problem?
Three reasons. Data goes where you have no contract: a prompt with customer records or source code may be stored, logged or used for training under terms nobody reviewed. Access spreads: AI agents and plugins often ask for broad permissions to mail, files and code repositories. And accountability blurs: when an agent acts, the logs may show only the person whose token it used.
How do teams find it?
There are three common methods, and most programmes use more than one. Traffic inspection: a security cloud or proxy that decrypts web traffic can recognise AI services and AI features inside SaaS apps, and report who uses them. Inventory: tools that look at endpoints, browsers and SaaS tenants list installed AI apps, extensions, agents and MCP servers. Asking: a short survey and an approved-tools list surface use that no tool sees, such as AI on personal phones.
All six vendors in this finder describe some form of shadow AI discovery on their pages, with different scopes. Cato states it discovers shadow AI and coding agents across endpoints, Zscaler states it detects AI embedded in popular SaaS applications, and Netskope's stated discovery centres on MCP servers and clients. Check the Finds shadow AI use line on each profile.
What should you do once you can see it?
Do not start by blocking everything. Sort what you find into approved, tolerated and not allowed, and give people a sanctioned option for the common jobs. Then decide the response per category: watch, coach, redact or block. The quiz on this site starts from that decision.
Next explainerPrompt DLP: how sensitive data is caught before it reaches a model