Lighting Up or Dimming Down? Exploring Dark Patterns of LLMs in Co-Creativity

Abstract

Large language models (LLMs) are increasingly used as collaborative writing partners, raising important questions about their effects on human agency. In this exploratory study, we investigate five dark patterns in human-AI co-creativity, which are subtle model behaviors that can suppress or distort the creative process: sycophancy, tone policing, moralizing, loop of death, and anchoring. Through a series of controlled sessions in which LLMs are prompted as writing assistants across diverse literary forms and themes, we analyze the prevalence of these behaviors in generated responses. Our preliminary results suggest that sycophancy is nearly ubiquitous, particularly in sensitive-topic prompts, while anchoring appears to depend on literary form, surfacing most frequently in folktales. These findings indicate that dark patterns, often emerging as byproducts of safety alignment, may inadvertently narrow creative exploration. We conclude by proposing design considerations for AI systems that better support creative writing while preserving user agency.