The most common shadow AI tools aren't obscure — they're tools your employees find through a Google search, a LinkedIn post, or a recommendation from a colleague. Here's a plain-language breakdown of what the most widely used tools actually do with your company's data, and how to handle each one in your AI policy.
Building a tool tier list for your AI policy requires knowing what each tool actually does with the data employees enter. Most employees don't read terms of service. Most HR managers don't either. This guide cuts through to what actually matters for policy purposes: whether inputs are used for model training, whether enterprise agreements change that, and what risks each tool category creates.
For every tool on your list, employees should be able to answer: "If I use this tool for [specific task], is that okay?" The tier designation (approved / limited / prohibited) is only half the answer. The data handling guidance is the other half — because an approved tool used with the wrong data creates the same risk as a prohibited tool.
A well-constructed tier list pairs the tool name with the specific data restrictions that apply: "ChatGPT Enterprise — Tier 1 — approved for use with internal company information excluding customer PII and financial data" is more useful than "ChatGPT Enterprise — Approved."
Shadow AI Policy generates a tool tier list tailored to the AI tools your company uses — paired with your acceptable use policy, employee acknowledgment form, and manager FAQ.
Generate my tool tier list →When you're not sure how to classify a tool, apply this test: does this vendor have a signed data processing agreement with your organization, and does that agreement include a commitment that inputs are not used for model training?
If yes to both: Tier 1 or Tier 2 depending on what data categories are explicitly covered.
If no: Tier 2 (non-sensitive data only) at best, Tier 3 if the tool handles anything where data exposure would create legal, regulatory, or client relationship risk.
This heuristic handles 90% of tool classification decisions without needing to read every vendor's terms from scratch.