Privacy of the extension
In short: your input never leaves your device through AImori. The check runs in the browser, on your device. This page describes what version 0.1.0 provably does – with evidence from the extension package.
On this page
No network access
- The extension contains no network code: no
fetch, noXMLHttpRequest, no WebSocket, nosendBeacon. An automated test in the extension's repository checks this on every change. - Its security policy also forbids every connection:
connect-src 'none'in the manifest. - No server, no account, no telemetry, no licence check. The reserved policy key
reportingUrlis never used.
Few permissions – and what for
AImori requests five fixed permissions, all for work inside your own browser:
storage: its own settings, counters and your own terms (as checksums only).scripting: works only together with the optional site permissions below, so only after your consent – for the notice or block on AI sites your organisation has not approved, the notice when copying, marking text copied from confidential sources and the form check. It does not extend the input check to further AI sites.offscreen: an invisible document of the extension in which text recognition for images and the local AI model run. Websites cannot open it.declarativeContent: the browser itself colours the icon on known AI sites (training traffic light). The extension does not learn which sites you visit.activeTab: only when you click the icon does the popup see which AI site is open. The address is not stored.
The extension checks input only on the supported AI sites. There are no host_permissions and no permission for history, downloads or bookmarks.
"permissions": ["storage", "scripting", "offscreen", "activeTab", "declarativeContent"]Optional, only after your click in the settings and revocable at any time:
- Further AI sites from the list: only for the notice or block page on AI sites your organisation has not approved. Input is not checked there.
- Confidential sources and form sites set by your organisation: text copied there gets a short-lived mark (a salted checksum only, 30 minutes); forms on these sites are checked locally before they are submitted.
- All https sites, only for the copy notice: if you copy text with sensitive data on a website, AImori shows a notice that AI sidebars and AI extensions are not checked. When you allow it, the browser asks clearly (roughly "read and change all your data on all websites") – needed because the notice can appear on any website. The copied text is only checked locally, never stored, never counted and never sent.
You can revoke optional access at any time in the extension's details under edge://extensions or chrome://extensions.
What is stored – and what is not
| Storage | Content | Duration |
|---|---|---|
| Settings | Your choices: language, country packs, global packs, action per data type | until you change them |
| Your own terms | only a salt and salted checksums, never the plain text | until you delete them |
| Statistics | Daily counters per AI site, category and decision – no text, values or times | 90 days |
| AI literacy record | Monthly counters per data type group, numbers only; can be turned off | 13 months |
| "Not sensitive" | per marked value only an HMAC checksum, never the value; at most 500 entries | 30 days |
Session (storage.session) | Account type per tab (business, personal, unknown), status of the local AI, an open follow-up prompt (only the type and number of matches), mark for text from confidential sources (salted checksum only) | until the browser closes; follow-up prompt at most 2 minutes, mark 30 minutes |
Never stored: your input, found values, excerpts, file names and visited addresses. Administrators cannot retrieve any content either – that function does not exist.
Values stay inside the checking part
Within the extension, only the type of data and the count move on, never the value. Hints show only "IBAN detected", never the IBAN. The hint belongs to the extension; the AI site cannot read it.
What AImori cannot prevent
AI sites can already read text while you type (for example for drafts). AImori checks before sending; what a site reads beforehand is not in the extension's hands. So it's best not to type sensitive data in the first place. More under Supported sites.
Managed devices
If your organisation distributes AImori, the popup shows "Managed by …"; hover over it to read: "No inputs are stored or reported; this extension has no network access." The organisation sets the rules but sees no content. Statistics stay on the device.
This website's privacy policy is a separate document: Privacy policy.
Local statistics
Deliberately minimal:
- Stored in the browser's local storage: daily counts only per AI site, category (e.g. financial data) and decision (redacted, cancelled, sent anyway, blocked, file notice, inactive, site blocked).
- No timestamps of individual events, no texts, excerpts, checksums of matches or file names. Deleted after 90 days.
- Visible in the popup (today, 30 days) and in the settings (30 days per category); "Reset statistics" is also there.
- On request, these counters produce the evidence report (totals only, to print or as an HTML file). Separately, the AI literacy record counts only monthly numbers per data type group.
- The counters never leave the device. The policy key
reportingUrlis only reserved for later reports; that requires a new version with its own security review, a fixed list of fields and totals only across several devices.