Microsoft Unveils LLM Security Scanner, Empowering Users to Detect Hidden Backdoors

Microsoft's groundbreaking research introduces a free security scanner to detect "sleeper agents" in open source Large Language Models (LLMs). This innovative tool allows users to verify their LLMs' safety, guarding against potentially malicious behaviors triggered by specific prompts. This proactive measure strengthens the safety and trustworthiness of open source AI.

• •Microsoft identified three key features to identify compromised LLMs.
• •The scanner helps users check if their Open Source LLMs are safe.
• •The research highlights the potential for malicious behavior through triggered prompts.