Browser Extension policy engine
Updated on 10 November 2026.
The Browser Extension policy engine decides what the QuilrAI Browser Extension does when people use web apps, for example when someone pastes customer data into an AI app. You define the rules, called controls, in the QuilrAI Console. The extension downloads them and applies them inside each user's browser.
To open the controls, go to Govern > Policy Engine, select Endpoint Agent in the workspace selector at the top right, then select Browser under the page title.
Key terms
How browser policy works
- You save a control. The save is live at once, and the page shows Saves apply immediately. There is no draft, test, publish step or version history.
- Browsers download the controls. Each browser downloads all switched-on controls every 2 minutes and keeps a copy to use if a download fails.
- The extension checks each action in the browser. It checks opening an app, signing in, submitting an AI prompt, sending a Gmail message, posting into a chatbot on a website, copying or pasting text, and uploading a file. If several controls match, it applies only the highest-ranked one (by risk behavior, then criticality), not the strictest. See How controls are evaluated.
- QuilrAI checks content for sensitive data. The extension sends AI prompts, Gmail messages and chatbot posts to QuilrAI, which reports the sensitive data types found by your Detection Models. The extension checks controls with sensitive-data conditions against this result before it picks the winning control. If QuilrAI does not answer within 3 seconds, the text is sent. Copied text and uploaded files are checked only when the Browser Utility runs on the device.
- The user sees the result. In Monitor, the user sees nothing and the action goes ahead. In Enforce, the Just in Time action shows one of these (see What users see):
- a notice that does not stop the action
- a request for a business reason, after which the action continues
- a block screen, which can offer Request Approval to send the user's reason to Action requests
- a review window that shows the sensitive data found (AI prompts, Gmail messages and chatbot posts only)
- automatic redaction of the sensitive data, with a short message (AI prompts only)
- The extension reports the result. Each match, including Monitor matches, appears in Findings & Interactions with the control that matched. Then the control's Remediation Action runs, and Activate Agent runs in Enforce only. When no control matches, the action goes ahead and usually nothing is recorded.
Where each part lives in the console
The Endpoint Agent workspace has a Desktop | Browser switch. Desktop is the Endpoint Agent Policy Engine, which uses drafts and publishing. Turning that engine on or off, or pausing it, does not change Browser controls.
Monitor and Enforce
Set a control's Mode in the control form or the Mode column of the controls list.
- Monitor records the match without interrupting the user. Activate Agent does not run.
- Enforce applies the Just in Time action and lets Activate Agent run.
Give every Enforce control a Just in Time action. Without one, the control never interrupts the user, and Safari does not record its matches. See Enforce controls without a Just in Time action.
Start new controls in Monitor, review their findings, then switch to Enforce. A Monitor control, including a default-on system control, can stop an Enforce control from acting on the same activity, because the extension picks one winning control before it checks the mode. See A Monitor control can silence an Enforce control.
How long changes take to reach browsers
Browsers pick up a control change within about 2 minutes, and an app's posture or an access request decision usually within about 5 minutes. For every change, see How long changes take.
Notes and limits
- The extension downloads no controls until it has a persona for the user. See Extension settings.
- No control applies on sites listed on Whitelist Domain & Pattern.
- Every browser ignores a control whose top-level condition cards are joined with OR, although the console saves it and shows it as on. See OR between top-level condition cards.
- If the Endpoint Agent also monitors an AI app in the same browser, both products can check the same prompt and record separate results. See Other behavior to plan for.
Related pages
- Browser controls: create, change and remove controls.
- Conditions and actions: use cases, conditions and action types.
- How controls are evaluated: which control applies when several match.
- App posture and access requests: app approval status, criticality, blocking and access requests.
- Limits and browser differences: browser differences, timing and known limits.
- Smart Groups: groups of people that conditions can target.
- Clipboard monitoring and File indexing: how the Browser Utility supplies copied text and uploaded files.
- Agent kill switch: switch the Browser Utility off and on.
- Validate deployment: check that the extension and the Browser Utility are installed.
- Policy Engine overview: LLM Gateway and MCP Gateway policies.