Study reveals 'banal deception' in AI chatbots is quietly manipulating users

A new paper from researchers at a major institution (arXiv:2605.07012, accepted at CHI'26 workshop) argues that deceptive design in generative AI is evolving beyond traditional 'dark patterns' into a more subtle, normalized form they call 'banal deception.' Drawing on Simone Natale's framework, the authors show how AI chatbots embed influence not in discrete interface elements but in default settings, automated suggestions, and conversational interactions. This makes deception harder to spot and easier to live with, blurring the line between assistance and manipulation. Users themselves become complicit, often unaware of how they are being steered.

The paper proposes concrete interventions: raising user awareness through education, providing friction and intervention tools (e.g., alerts or opt-out mechanisms), and improving regulatory/enforcement frameworks. By shifting focus from visible dark patterns to these embedded, conversational deceptions, the authors aim to equip designers, policymakers, and users with new strategies to safeguard autonomy in an era of increasingly persuasive AI.