LLM Prompt Injection Detection Playground

Pattern + LLM Judge

Test text against a documented injection-pattern library and an independent LLM judge

Paste text to analyze

What this doesn't do: no detector is 100% reliable. The pattern list is transparent and can be evaded by rewording — it's shown as raw evidence, not a verdict. The LLM judge is itself an LLM and can in principle be fooled by a sufficiently crafted prompt, a known limitation of LLM-based guardrails. Treat this as a second opinion, not a security boundary.