Research
Reasoning Made GUI Agents More Resistant to Default Nudges and More Vulnerable to Social Ones
arXiv 2609.19843 (17 Sep 2026) ran a randomized online shopping experiment with 3,600 GUI agents and 21,600 simulations across six frontier models from three providers, testing whether agents fall for the same interface nudges designed to steer humans. Agents were susceptible to both automatic (Type 1) and reflective (Type 2) nudges, and reasoning configuration moved the two in opposite directions: extended reasoning cut susceptibility to automatic default nudges while raising it to reflective social-influence nudges. More reasoning did not produce a more robust agent, it changed which manipulation worked.
↳ Follow the thread