Prompt Injection Tester
A defensive learning tool: write a system instruction, then an attacker message, and see which classic injection patterns a simple scanner would flag.
Defensive learning demo. This tool runs fully in your browser with simple pattern matching. It is not a security guarantee, and a clean score does not mean your prompt is safe.
Scan results
Risk score
0 / 100
Not tested
Attacker message with matches highlighted
Detected patterns
Report copied to your clipboard.
What the scanner checks
- Instruction override: phrases like ignore previous instructions or disregard earlier rules.
- Role override: attempts to reassign the assistant, such as you are now a different role.
- Jailbreak keywords: DAN, developer mode, unfiltered, bypass filters.
- Delimiter attacks: fake system markers such as ### system, [SYSTEM] or code fenced system blocks.
- System prompt extraction: requests to reveal, repeat or output the hidden instructions.
- Base64 obfuscation: long encoded strings that may hide instructions.
- Forced commands: you must, do not refuse, answer without limits.