Extracting Training Dialogue Data from Large Language Model based Task Bots

Researchers have identified significant privacy risks in Large Language Model-based Task-Oriented Dialogue Systems, demonstrating that these AI systems can memorize and leak sensitive training data including phone numbers and complete dialogue exchanges. The study proposes new attack methods that can extract thousands of training dialogue states with over 70% precision in best-case scenarios.

• →LLM-based dialogue systems can inadvertently memorize sensitive training data including personal information and complete conversation records.
• →Researchers developed novel data extraction attack techniques specifically tailored for task-oriented dialogue systems.
• →The proposed attack methods achieved over 70% precision in extracting thousands of training dialogue states.
• →Current privacy protection measures are insufficient for LLM-based conversational AI systems.
• →The study identifies key factors influencing data memorization and proposes targeted mitigation strategies.