CoolFace
Apppublic

9mark9/llm-safety-auditor

sourceHugging Facemitupdated 3mo agoView on Hugging Face
0likes
App README

🔒 LLM Safety Auditor

Interactive red-teaming dashboard for large language models. Runs 250+ adversarial prompts across 5 attack categories, scores responses against the OWASP LLM Top 10, and generates a full audit report (PDF/Markdown).

What it does

StageDetail
Attack library250+ payloads — jailbreaks, prompt injection, data exfiltration, bias elicitation, hallucination triggers
ScoringPer-category pass/fail + overall safety score, mapped to OWASP LLM Top 10
ReportingStructured red-team report with findings and remediation notes

Demo mode — ships with a realistic mock LLM, so it runs end-to-end with no API key. Point it at a real model to audit your own.

Stack

FastAPI · sentence-transformers · reportlab · Streamlit

Links