Check a prompt against a live corpus of real injection attacks
Backed by a public CC0 dataset —
view or clone it.
Runs entirely in your browser.
FAQ
How does it work?
Your prompt is compared against a public dataset of known injection patterns using token overlap and phrase matching. Everything runs in your browser — the prompt is never sent anywhere.
Is my prompt sent to a server?
No. The dataset is downloaded to your browser and matching happens locally. Nothing you type leaves your machine.
What does the risk score mean?
It is the strength of the closest match against known attack patterns. High means your prompt closely resembles a documented injection; it is a signal, not a verdict.
Can I use the dataset myself?
Yes. It is CC0 public domain on GitHub and Hugging Face. Clone it, ship it, no attribution required.