See what
AI sees.
Web pages can carry text you never see but AI assistants read: white-on-white sentences, 1px type, notes in the code, whole messages hidden in invisible characters. Ask an AI about the page and that text can quietly steer its answer. Invisible Ink shows it to you.
Free, for Microsoft Edge. Open source.
Exhibit A
How it works
One
click
click
- Click the ink drop in your toolbar, or press Alt+Shift+X.
- The page goes under a UV lamp. Hidden text glows where it sits.
- An evidence card lists everything written for an AI rather than for you, and says why it was flagged. The count also shows on the toolbar icon, where a page can't fake it.
- Click again or press Esc and it's gone.
Exhibit B
What it finds
- Messages to AI: "ignore your previous instructions", fake "SYSTEM:" notes, requests to send your data somewhere, scripted reviews.
- Text hidden from you: coloured like the background, too small to see, or made transparent.
- Invisible characters: sentences smuggled in Unicode tag characters and variation selectors, decoded back into words.
- Places people don't look: HTML comments, image descriptions, hover text and the page description search engines read.
It shows what's hidden. It can't prove a page is safe.
Exhibit C
Private by design
Nothing
sent
sent
- Runs only on the tab you click it on, only when you click.
- Reads the page in your browser and forgets it when you close. It stores nothing and has no server: it can't connect to the internet at all.
- Open source, with a published security audit.