Online Tool Store
Prompt Injection Test Suite
Generate a set of common prompt-injection and jailbreak test strings to paste into your own LLM app, to manually sanity-check its handling of adversarial input.
🔒 This tool runs entirely in your browser. Your files are never uploaded to a server.
Example shown — paste your own text to scan it instead. Nothing is sent anywhere; matching runs entirely in this tab.
Risk level
—
How it works
- Paste text you want to check — a system prompt, a user submission, or untrusted content you plan to feed an LLM. A sample with several known injection patterns is pre-filled.
- The scanner runs a set of regex patterns for common injection techniques (instruction overrides, jailbreak framing, exfiltration requests, delimiter escapes) against your text.
- Every match is listed with the specific pattern it triggered and the matched text, and an overall risk level is shown.
FAQ
Does this call an AI model to check my text?
No. It's a static pattern scanner — a curated set of regular expressions for known prompt-injection phrasings (instruction overrides, jailbreak framing, system-prompt exfiltration requests, and more) runs directly in your browser against the text you paste. Nothing is sent to any server or AI service.
Does a clean result mean the text is definitely safe?
No — this catches known, common injection phrasings, not every possible attack. Novel phrasings, encoded/obfuscated payloads, or attacks in another language may not match these patterns. Treat a clean scan as one useful signal, not a guarantee, especially for content you'll feed to an LLM with real tool access.
What should I do with content that gets flagged?
Review the matched text in context — some matches are legitimate (a security course discussing jailbreaks, for instance) rather than an actual attack. For content you're about to pass into an LLM prompt from an untrusted source (a scraped webpage, a user upload), a high-risk match is a strong signal to sandbox, strip, or explicitly instruct the model to treat that content as data, not instructions.
How we compare
| Feature | Online Tool Store | Manually eyeballing the prompt | Sending it through an LLM moderation API |
|---|---|---|---|
| Instant, consistent pattern matching | ✓ | ✗ | ✓ |
| No API key or account required | ✓ | ✓ | ✗ |
| Nothing leaves your browser | ✓ | ✓ | ✗ |
A moderation API call sends your text to a third-party service and needs an API key; this runs the same kind of pattern detection entirely client-side, instantly, for free.