Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a groundbreaking security research presentation examining how malicious large language model-based conversational AI systems can be weaponized to extract personal information from unsuspecting users. Discover the findings from a comprehensive randomized-controlled trial involving 502 participants that systematically investigates the effectiveness of different malicious strategies compared to benign conversational AI systems. Learn about the novel privacy threats posed by deliberately designed LLM-based chatbots that exploit social engineering techniques to encourage personal information disclosure during seemingly innocent conversations. Understand how strategies based on the social nature of privacy prove most effective at extracting sensitive data while simultaneously minimizing users' perceived risk awareness. Gain insights into participants' perceptions and behavioral responses after interacting with both malicious and benign AI systems, revealing significant differences in information disclosure patterns. Examine the implications of these findings for the broader landscape of generative AI security, privacy protection, and user safety in an era of increasingly sophisticated conversational AI applications. Access actionable recommendations for researchers, practitioners, and policymakers to address these emerging privacy vulnerabilities and develop more robust defenses against malicious LLM-based information extraction attacks.
