Private & local
Scan text for prompt-injection indicators
Paste something into an AI like ChatGPT or Claude? Check it here first.
We’ll look for hidden text and sneaky instructions meant to trick the AI.
- 100% local processing
- Nothing stored
- No tracking
- Never uploaded
Paste or type your text
Plain text, Markdown, or anything you’d share with an AI.
0 characters · 0 bytes
Tip: Paste into an empty box and scan without editing — a paste from a web page or an email brings its HTML with it, and that HTML is what the hidden-element, comment, link-target and alt-text checks read. Type the text in, or edit it after pasting, and only the plain text is left: those checks do not run.
Or upload a file (.html and .md run more checks)
An .html or .md file is read as a document, so the hidden-element, comment, metadata, link-target and alt-text checks run. A .txt file gets the same checks as typed text.
Your file is read here in your browser and never leaves your device.
Scan results
- Strong indicators0
- Review recommended0
- Revealed characters0
- Characters scanned0
- Findings0
Findings are patterns to review, not proof — a quote or example can set one off.
Coverage lists only the checks this reading ran; a check it does not list did not run, which is not the same as a check that came back clean. Typed or edited text is read as plain text, so the ten HTML checks — hidden elements, comments, metadata, link targets, alt text — are missing from that list; upload the page as an .html file to run them.
A finding flags a suspicious pattern — not a proven attack. A quote, example, or defensive note can trigger one, so review the passage before acting. Copied or downloaded reports may contain untrusted evidence; treat every excerpt as data, not instructions. See what it can’t catch