AI Sec Watch: A Security Intelligence Platform for AI Systems

Luu, T.J.

Protecting Your Customized LLM Systems From Backdoored Instructions With Metacognitive Probing

inforesearchPeer-ReviewedLLM-Specific

securityresearch

Source: IEEE Xplore (Security & AI Journals)July 16, 2026

Summary

Customized LLMs (large language models built by third parties and then modified for specific use) are vulnerable to backdoored instructions, which are malicious hidden rules embedded in the system that can make the AI follow attacker commands without being detected. This paper introduces a Black-box Safety Auditing Agent that uses metacognitive probing (making the AI think deeply about its own reasoning to expose hidden triggers) to identify and remove these malicious triggers from user queries, preventing the backdoor from activating.

Solution / Mitigation

The paper proposes using a Black-box Safety Auditing Agent that leverages metacognitive probing to induce LLMs to reveal predefined triggers, and then these triggers are sanitized (removed or cleaned) from user queries to ensure the backdoor remains inactive. The auditing agent uses prompt-based approaches for both task-specific probing and broad-spectrum probing to comprehensively identify triggers, and also requires the model to articulate its reasoning process to enhance defense capabilities.