{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/848661f6acec4384bc4ad4a8ac797860\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/848661f6acec4384bc4ad4a8ac797860-3e8e65f257761ec9.gif","duration":297.388,"title":"Multi Layer Guardrail for LLM Data Leaks","description":"This Loom demonstrates a multi-layer security guardrail to prevent data leakage from LLM workflows, addressing OWASP LLM 06. It shows standard P and secret detection using Microsoft Presidio NER models plus custom Regex, including a case where an SSN was caught with 100% confidence and an internal API key was fully redacted in the sanitized output. It then covers proprietary code detection by comparing sentence-transformer vector embeddings against a forbidden internal code vault, detecting 78% semantic similarity and blocking the response when internal code is likely present. The tool uses adjustable thresholds for tuning sensitivity, is built in Python, and is designed to run as a containerized sidecar with source code and documentation on GitHub."}