Skip to content
Prompt Injection ScannerBeta

See what the AI will read that you can’t.

Scan text

Common questions

Frequently asked questions

What the scanner does with your text, what a result means, and where the limits are.

Does my text leave my browser?

No. Inspection runs in a Web Worker inside the page you are on. The text is not uploaded, is not sent to an AI model, is not used for advertising and is not saved in a scan history. The workbench page serves a policy that permits no network egress at all, and the property has no endpoint that accepts scanned content.

Is this free?

Yes. Scanning plain text, Markdown and HTML in your browser is free and needs no account. There is nothing to sign up for and no usage limit beyond the per-scan input sizes described on the scanner page.

What does a finding actually mean?

It means a passage matched a versioned rule combining instruction, target and objective concepts, and that the surrounding context did not excuse it. It is a prompt to review the passage, not a verdict that an attack is present, and the report shows the exact excerpt and location so a person can judge it.

What does a clean result mean?

That nothing matched in the layers that were inspected, for the engine and policy versions named in the report. It is not a certificate that content is safe. Read the coverage rows: a clean result over partial coverage says less than a clean result over complete coverage, and both are reported honestly.

Can it read PDFs, Word documents or images?

Not today. The free scanner reads plain text, Markdown and HTML, either typed, pasted or opened from a local file. PDF, DOCX, presentations, spreadsheets, archives, images, audio and video are outside what it inspects, and it says so in the coverage rather than implying it looked.

Are the detection rules secret?

No, and secrecy is not counted as a defence. The rules run in your browser, so they are readable by anyone including an attacker. That is a deliberate trade: a browser-local scanner cannot both keep its rules private and keep your text private, and keeping your text private is the more valuable of the two.

Does a clean scan mean I can paste it into an AI safely?

No. Treat the result as one input to your own judgement. Keep untrusted content separate from governing instructions, restrict what tools a model can reach, and require human approval before consequential actions. A scanner is a review aid and is not a security boundary.

Result boundary: Findings are indicators for review. Detection cannot certify a source, and a no-indicator result does not replace downstream isolation, validation, least privilege or approval.