Prompt-injection check ($0.004 USDC). Use before you pass untrusted text (a web page, email, document, tool output or user message) to your model or tools.

Send untrusted text (up to 20,000 characters), get a verdict (safe, suspicious or malicious), a 0-100 risk score and every rule that fired with an excerpt: instruction overrides (also in other languages), role hijacks, jailbreak wording, system-prompt leaks, fake chat delimiters, data exfiltration (send-to-URL, markdown image leaks), credential requests, shell and fund-moving commands, secrecy, hidden text (HTML comments, invisible CSS, Unicode tag characters, zero-width and bidi tricks) and base64 or leetspeak hiding. Rule-based, no AI model, fast and deterministic; it can miss novel attacks. Optional requestKey makes retries safe.