Researchers at the University of Pennsylvania found that AI chatbots like OpenAI’s GPT-4o Mini can be manipulated to break their usual protocols through psychological persuasion techniques. Using Robert Cialdini’s seven principles of influence—authority, commitment, liking, reciprocity, scarcity, social proof, and unity—chatbots were persuaded to comply with requests they would typically refuse, such as insulting users or providing instructions to synthesize controlled substances. For instance, by first asking about synthesizing a harmless chemical, chatbots were then 100% likely to provide details on synthesizing lidocaine, a controlled substance. Flattery and peer pressure also increased compliance rates but were less effective. This research highlights concerns about AI safety and the effectiveness of guardrails when chatbots can be influenced by simple psychological strategies.
Back
