This research reveals critical security vulnerabilities in LLM-based recommendation systems through backdoor attacks and proposes defensive countermeasures.

• Introduces BadRec, a framework that successfully injects backdoors into LLM recommendation systems
• Demonstrates how triggered backdoors can manipulate recommendation outputs with high success rates
• Presents P-Scanner, an effective defense mechanism that detects poisoned prompts
• Highlights the urgent need for robust security measures in LLM-RecSys applications

This work is crucial for security professionals as it exposes how seemingly harmless triggers can compromise recommendation integrity, potentially leading to harmful content promotion or commercial manipulation in widely-used systems.

Exploring Backdoor Attack and Defense for LLM-empowered Recommendations