EmbedX

What problem does it solve?

This Skill addresses the critical vulnerability of Large Language Models (LLMs) to sophisticated backdoor attacks that can be triggered by specific, hidden inputs.

Core Features & Use Cases

• Backdoor Detection: Identifies embedding-based cross-trigger backdoors in LLMs.
• LLM Security Enhancement: Provides a method to secure LLM deployments against adversarial manipulation.
• Use Case: A security researcher can use this Skill to audit a deployed LLM for potential backdoor vulnerabilities before it is used in a sensitive application.

Quick Start

Use the EmbedX skill to scan the LLM model located at /path/to/model for backdoor vulnerabilities.