Skip to content
The most popular shadow AI tools are general-purpose chatbots such as ChatGPT, Claude, Gemini and Perplexity. Close behind are workplace assistants like Microsoft Copilot, AI features built into everyday apps such as Notion AI and coding assistants.
Employees adopt these tools without approval because they are quick to access, free or low-cost at entry level and immediately useful. The specific tools, however, matter far less than whether security teams can see them and understand how they’re being used.
The Main Categories Of Shadow AI Tools
Most unsanctioned AI use falls into four categories. Each one enters the organization uniquely, supports different kinds of work and puts separate types of data at risk.
General-Purpose Chatbots
General-purpose chatbots are the most widely used shadow AI tools. Many are available for free in a browser, so employees can start using them in seconds with a personal email address. Common uses include drafting, summarizing, translating and research.
The risk lies in what employees paste in. Summarizing a contract or analyzing a spreadsheet means sharing the content itself, and on personal accounts, that data may sit outside the organization’s control.
Workplace Assistants
Workplace assistants are AI tools built for business productivity. They become shadow AI when employees use them through personal accounts rather than the organization’s approved deployment.
Company versions typically include data protection commitments and admin controls that personal accounts lack, meaning the same prompt carries very different risk depending on where it goes.
Embedded AI Features
Embedded AI features are functions built into software the organization already uses. Vendors often switch them on through routine updates, so shadow AI can appear inside an approved app without anyone installing anything.
This makes them the hardest category to spot. Meeting transcripts, internal documents and customer notes can all pass through without a dedicated approval decision.
Coding Assistants
Coding assistants are AI tools that help developers write, explain and debug code. They include browser-based chatbots and plugins for integrated development environments (IDEs).
Developers adopt them to speed up repetitive work, but code often contains more than logic. Proprietary source code, application programming interface (API) keys and internal architecture details can all end up in prompts, exposing intellectual property and creating openings for threat actors.
Why Visibility Matters More Than The Tool
New AI tools launch constantly, so any blocklist of named services falls out of date quickly. The risk remains regardless of which tool an employee chooses, because sensitive data leaves the organization either way. The real challenge is that most businesses can’t detect shadow AI across every tool at once.
Common security approaches each see only part of the picture. Network-layer steps, such as a cloud access security broker, can’t read conversational data moving through encrypted browser sessions to legitimate AI websites. Browser extensions only cover managed browsers, so they miss personal devices and AI embedded in approved apps. API gateways only monitor the enterprise AI deployments an organization has sanctioned. This fragmented view is how the shadow AI economy grows unnoticed.
How BlackFog Provides Visibility Across Every AI Tool
ADX Vision addresses shadow AI at the endpoint. It runs directly on the device and inspects AI-bound data before it’s encrypted or transmitted, across supported applications, browsers or agents. Because it isn’t tied to any AI provider, it can apply consistent protection to chatbots, embedded AI features and coding plugins, including newly introduced tools.
Organizations can discover which AI tools employees use, prevent sensitive data from reaching unapproved services and enforce one AI usage policy on every device, including those off the corporate network.
