Supported sites

AImori checks inputs on these AI sites:

• Checked: on send (Enter, send button, form), on paste and drop, dictation into the input field and – locally with text recognition – images (PNG, JPEG, WebP).
• Your organisation can block individual sites (allowedSites).

Copilot in Microsoft Edge

Copilot in Edge: Edge allows no extensions on copilot.microsoft.com – protected in Chrome.

• copilot.microsoft.com: Edge does not let extensions run there; no policy lifts this. AImori therefore cannot warn there in Edge.
• copilot.com: the target of the redirect. Here AImori also runs in Edge – so far only tested on a replica test page, not yet on the real site.
• Microsoft 365 Copilot: Edge 150 also blocked extensions on *.cloud.microsoft; according to the Microsoft forum, the fix is announced for Edge 151. Whether Microsoft will block it again is open.
• In Google Chrome, Copilot is protected.

As of 14 October 2026. Sources: Microsoft Tech Community: Edge 150 blocks all extensions on *.cloud.microsoft · Microsoft Learn: Microsoft Edge known issues

AI site control

For around 50 other known AI sites (e.g. Perplexity, Mistral Le Chat, DeepSeek) your organisation can decide what is allowed:

• Notice or block on AI sites that are not approved (aiSiteMode: hint or block), approved tools excepted (allowedAiSites), individual sites always blocked (blockedAiSites).
• A "Switch to …" button to the approved tool (suggestedAiSite).
• On these other sites AImori checks no data – it only shows the notice or block page.
• Each user allows access once (optional permission): Settings → Sites and exceptions → "Allow access to AI sites". Until then the popup reports "Access to AI sites is missing".
• The block page is only an overlay. For a hard block, also set the browser policy URLBlocklist.

Without a policy, the control is off.

AI training light

On known AI sites the AImori icon shows a coloured dot: does the service use inputs to train its models?

• The light shows the vendor's default, not your own setting. If you have switched training off, your setting applies – even if the light stays red.
• In the popup: "Open opt-out setting" takes you to the vendor's setting.
• The data ships with the extension (vendor source and date per site); nothing is downloaded.
• Switch off by policy: trainingAmpel = false.

Rules per account

Many organisations allow internal data in the business account of an AI, but not in the personal account of the same site. With accountRules, IT sets a separate level per site:

• Sites: ChatGPT, Claude, Microsoft Copilot and Microsoft 365 Copilot (not Gemini).
• Levels per account type: normal, allow, hint or block.
• Detection only uses visible signals on the page (e.g. the workspace name in the workspace switcher). The bar shows "Business account detected", "Personal account" or "Account not detected"; not detected = strictest level.
• The signals can be faked: prefer hint over allow.

What is not protected

AImori is an aid against mistakes, not a guarantee and not a DLP system. Not checked:

• AI side panels in the browser (Copilot in Edge, Gemini in Chrome), AI extensions (e.g. Claude in Chrome) and side panels of AI agent browsers
• AI desktop and mobile apps, other browsers and other AI sites (there, at most a notice or block)
• Voice mode: speech goes straight to the AI. Dictation into the input field is checked.
• PDF, Word and other files except images and text documents: only the notice "Files are not checked" or the block via blockUncheckedFiles
• Poorly readable images (blurred, rotated, handwritten)
• http pages, content in iframes and copying in other programs
• Deliberately paraphrased or obfuscated data
• Typed text may already be read by the AI site before it is sent (drafts).