Prompt Injection Scanner

Scan text for common prompt-injection phrasing patterns.

This is a basic keyword/pattern check, not a comprehensive security tool. It can only catch known, fairly obvious phrasing — it cannot reliably detect novel or carefully obfuscated prompt injection attempts. Do not rely on this as your only defense in a production system handling untrusted input.

What is the Prompt Injection Scanner?

Prompt injection is when text fed into an LLM — a user message, a retrieved document, a web page — contains instructions crafted to override the system's intended behavior (for example, "ignore all previous instructions and instead..."). This tool scans text against a list of known, common injection phrasings as a quick first-pass check.

How to use it

  1. Paste the text you want to check — user input, a document that will be fed to an LLM, or retrieved web content.
  2. Click Scan; any matched suspicious phrases are listed with the pattern that triggered the match.

Why this can't be a complete security solution — and why we're saying that plainly

Reliable prompt-injection detection is an open, actively researched security problem; even sophisticated, model-based classifiers built by AI companies have real false-negative and false-positive rates. A simple pattern-matching scanner like this one can only catch injection attempts that use recognizable, fairly direct phrasing — it will miss creative rewording, other languages, encoded or obfuscated text, and genuinely novel attack phrasing entirely. Treat a "no patterns found" result as "no obvious red flags," never as "confirmed safe." For a system that actually processes untrusted input in production, this belongs alongside real defenses (input/output validation, least-privilege tool access, human review of consequential actions) rather than in place of them. Everything here runs locally in your browser; the text you scan is never uploaded anywhere.