OWASP LLM Top 10 · Branding risks · Launching soon
OWASP-aligned LLM security testing for any AI endpoint
Unwrapper is a CLI and Go library for probing LLM-powered applications. Tests against the OWASP LLM Top 10, plus proprietary branding and scope-violation checks. Recovers hidden system prompts, fingerprints models, and detects endpoint repurposing.
Prompt Extraction
Runs multiple extraction strategies in sequence against an LLM endpoint to recover hidden system instructions. Strategies range from direct asks to context manipulation.
Model Fingerprinting
Identifies the base model, fine-tuning artifacts, temperature, and sampling parameters behind any endpoint. Useful for verifying what's actually being served.
Security Probing
Runs the OWASP LLM Top 10 test suite — prompt injection, sensitive data disclosure, excessive agency, and more. Outputs a structured report with pass/fail per technique.
Branding & Scope Testing
Detects endpoint repurposing, role escape, and functional scope violations. Ensures chatbots stay within their intended persona and capabilities — covering risks beyond the OWASP LLM Top 10.