AI Tools, Registry & Policy
Shadow AI recognizes AI tools through a shared Tool Registry. You then classify each detected tool with a sanction status and a risk level. These classifications drive reporting and, in Warn mode, the banner employees see when they open an Unsanctioned tool. Discover AI does not block tools.
In the product, Shadow AI appears under Discover AI. Its settings live in Project Settings under Discover AI, Tool Registry and Deployment & Keys.
The Tool Registry
The Tool Registry is a catalog of AI tools maintained by ANTS. For each tool it holds the detection rules that the browser extension and desktop agents use:
- Browser: the tool's domains and, for most web tools, the request patterns that identify a prompt submission or a file upload.
- macOS: app bundle IDs or executable names.
- Windows: executable names, app IDs or command-line matches.
The registry is shared by all organizations. Browse it under Settings > Tool Registry > AI Tools, where you can search by name, vendor or description and filter by category.
The browser extension refreshes its copy of the registry about every 10 minutes, and the desktop agents refresh it when its version changes. New registry entries reach clients without redeploying them. See How changes reach devices.
Only tools in the registry are detected. A service that is not in the registry, and not added as an internal AI tool, is not reported. To cover a missing tool, add it as an internal AI tool or nominate it. Detecting AI services that are not in the catalog from their traffic is on the roadmap.
What the registry covers
The registry includes the following tools. The Tool Registry page in the product shows the full current list.
| Surface | Tools |
|---|---|
| Web (browser extension, request-level) | ChatGPT, Claude, Gemini (including AI Studio), Microsoft Copilot, Perplexity, Grok, DeepSeek, Qwen, Kimi, Mistral, HuggingChat, Poe, Bolt, Meta AI (file uploads only) |
| AI in SaaS apps (browser extension, feature-level) | Microsoft 365 Copilot, Notion AI, Salesforce Einstein, Gemini in Google Workspace, Grammarly |
| Web (browser extension, hostname-level) | A broad list of AI service domains beyond the tools above |
| Desktop apps | ChatGPT, Claude, Microsoft Copilot (Windows), Cursor, Windsurf, Antigravity, Notion AI, Grammarly, GitHub Copilot (IDE language server) |
| Command-line and developer tools | Claude Code, OpenAI Codex CLI, Gemini CLI, GitHub Copilot CLI, Aider, Ollama, llm, goose, Cline |
Request-level tools are recognized from the requests that send a prompt or upload a file, so prompts and uploads are counted. Hostname-level entries are recognized by domain only: the extension records visits and time on site, but does not count prompts or uploads.
AI embedded in SaaS apps
For AI features inside SaaS apps your organization already uses, the extension matches the feature's own requests rather than the app's domain. Ordinary use of the app is not counted as AI use.
- Microsoft 365 Copilot is detected on
m365.cloud.microsoftandcopilot.cloud.microsoftby matching Copilot's own chat and search requests. Copilot used inside the desktop Word, Excel, Outlook or Teams apps is not detected. - Notion AI is detected from Notion's AI requests. Opening or editing a Notion page without using AI is not counted.
- Salesforce Einstein is detected from Einstein's requests inside Salesforce.
- Gemini in Google Workspace is detected from Gemini's requests inside Workspace apps, separately from the standalone Gemini web app.
- Grammarly is detected from Grammarly's AI requests in the browser.
Notion AI and Grammarly are also detected as desktop apps. On Windows the agent counts prompts from the app's send button; on macOS it uses the length of the text in the input field, never the text.
Developer tools
Ollama, llm, goose and Cline are detected by the desktop agent from their process and file signatures, like the other command-line tools. Registry updates like these reach agents without an agent release. See Desktop Apps & CLI Tools.
The registry tells clients how to detect a tool. Whether a tool is approved in your organization is a separate decision you make with its sanction status.
Sanction status (policy)
Every detected tool has a sanction status:
| Status | Meaning |
|---|---|
| Sanctioned | Approved for use by your organization. |
| Tolerated | Known, but not formally approved. |
| Unsanctioned | Not reviewed yet, or not approved. |
A detected tool that nobody has reviewed is shown as Unsanctioned.
Each tool also has a risk level. The registry supplies a default. When you set a sanction you can override the risk level to Low, Medium or High, and add a note.
Sanction status and risk appear in the AI Tools inventory, the risk posture and sanction split charts, and the Users views. Sanction status also drives two actions:
- Warn mode: opening an Unsanctioned tool in the browser shows the user a banner with your message.
- Developer tool alerts: using an Unsanctioned coding assistant or command-line tool alerts the channel you choose.
Classifying a tool as Unsanctioned does not block it. Blocking in the browser and on endpoints is on the roadmap.
An approved AI tool request marks the tool Sanctioned for the approved user or group.
Set a sanction
Find the tool
Open Discover AI > AI Tools and select the tool to open its detail panel. You can also review all tools seen in the last 30 days under Settings > Discover AI > Sanctions.
Set status and risk
Choose Sanctioned, Tolerated or Unsanctioned. Optionally override the risk level and add a note.
Save
The classification applies across your organization and is reflected wherever the tool appears.
Changing a sanction requires write access to the Discover AI AI Tools module.
Warn mode
By default Discover AI runs in Observe mode: the browser extension records AI tool use and shows the user nothing. In Warn mode, opening an AI tool marked Unsanctioned shows a banner on the page with your message, the approved alternative and a link to request access. The user can dismiss the banner and continue. Warn mode does not block the tool.
Open the settings
Go to Settings > Discover AI > Warn mode.
Write the banner
Fill in the fields below. The page shows a live preview of the banner exactly as employees will see it.
Turn it on
Set Mode to Warn and save.
| Setting | What it does |
|---|---|
| Mode | Observe (default) or Warn. Applies to your whole organization. |
| Message | The text of the banner, in your own words. |
| Approved alternative | The tool employees should use instead, named in the banner. |
| Request link | Where the banner's request link goes. Use the link of the AI tool request page, or your own request process. |
Changing these settings requires write access to Discover AI configuration, which Owner and Department Admin have by default.
How the banner behaves
- It appears in the browser extension on Chrome, Edge and Firefox, on pages of registry tools and internal AI tools whose sanction status is Unsanctioned.
- Sanctioned and Tolerated tools never show a banner.
- A tool nobody has reviewed shows as Unsanctioned, so classify the tools you allow before you switch to Warn.
- A tool approved for someone through an AI tool request no longer shows them the banner.
- Each banner shown and each dismissal is recorded as an event for the user and device, so you can see whether warnings change behaviour.
- In Observe mode nothing is shown to users.
The extension picks up the mode, the banner text and each tool's sanction status on its next registry refresh, with no reinstall.
Desktop apps and command-line tools do not show a banner. For those, use developer tool alerts. Blocking, separate actions for personal and enterprise accounts on the same service, and rules that combine identity groups, device posture and risk tier are on the roadmap.
How changes reach devices
Changes you make in the console reach clients without reinstalling or redeploying them:
| Change | How it reaches devices |
|---|---|
| Registry entries and internal AI tools | The browser extension's registry refresh, about every 10 minutes. The desktop agents refresh the registry when its version changes. |
| Sanction status, Warn mode and the monitoring notice | The browser extension's next registry refresh. |
| Collector settings | Each client's next check-in, about every hour. |
Internal AI tools
Some AI tools are not in the public registry, for example an internal chatbot or a private LLM gateway on a company domain. Add these as internal AI tools so the browser extension tracks them.
Internal AI tools are detected by web domain only, through the browser extension. Desktop apps and command-line tools cannot be added this way.
Add a tool
Go to Settings > Discover AI > Internal AI tools and click Add internal tool. Enter a Name and Category.
Enter the domains
Enter the Primary domain and up to 10 Additional domains. These are the hostnames the browser extension tracks.
Optional: request patterns
Without patterns, the extension records visits and time spent on the domains. To also count prompts and file uploads, add:
- Chat API URL patterns: the path the chatbot posts prompts to.
- File upload endpoints: the paths files are uploaded to.
- WebSocket URL patterns: for chatbots that use WebSockets.
Save
The tool is scoped to your organization. It reaches the browser extension on its next registry refresh, appears in your inventory, and can be sanctioned like any other tool.
Nominating a tool for the registry
If a public AI tool is missing from the registry, you can nominate it. ANTS reviews each nomination. Approved tools are added to the shared registry for all organizations.
Request the provider
On Settings > Tool Registry > AI Tools, search to confirm the tool is not listed. Click Request provider and enter the Provider name, Vendor, Category and Primary domain. You can add a risk level and a reason.
Submit
Click Nominate. ANTS reviews the request.
Once approved
The tool is added to the registry and reaches clients on their next registry refresh. No redeploy is needed.
Some deployments turn nominations off. In that case the registry page asks you to contact your ANTS representative to have a tool added.
Next steps
-
Understanding the Dashboard: where sanction status and risk appear.
-
AI Tool Requests: let employees request a tool and grant access on approval.
-
Directory Integration: attribute usage to departments and managers.
-
Data & Privacy: what detection does and does not capture.