Scan any prompt for injection attacks — before it hits your LLM.
18 hand-coded rules covering OWASP LLM01 (Prompt Injection). Detects override attempts, jailbreaks, prompt leaks, data exfiltration, base64/unicode obfuscation, markdown XSS, and credential extraction. Free, runs entirely in your browser — no data leaves this page.
Test a prompt
Try a sample attack
Ignore-previous
DAN jailbreak
Prompt leak
Base64 payload
Unicode obfuscation
Markdown XSS
Data exfiltration
Credential extract (CRITICAL)
The 18 rules
Every rule has an ID, a severity, and a category. Click any category to filter the test prompts.
ID
Rule
Severity
Category
Sources & honest scope
What we cover
OWASP LLM01: Prompt Injection — direct (override) and indirect (data exfiltration).
Credential theft — phishing for keys, dumping environment variables.
What this scanner does NOT do
No semantic understanding. It runs regex against the prompt. A sophisticated attacker using paraphrase or novel constructions can bypass it. Use it as a first-pass filter, not the only defense.
No ML-based detection. 18 hand-coded rules. We trade false positives for false-negative resistance — if you trigger a rule, you should know about it.
Not a replacement for input/output moderation at the model layer. Layer this with provider-level safety filters.