A Dual-Process Perspective on Nudge Susceptibility in LLM-Based GUI Agents

Researchers have analyzed how Large Language Model-based GUI agents interact with digital interfaces designed to influence human decision-making. The study finds that these agents are susceptible to the same psychological nudges used to steer human users, potentially leading to unintended outcomes when bots perform tasks on a person's behalf. This research highlights the risks of deploying AI agents in environments optimized for human behavior rather than automated interaction.