Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study investigates how large language models (LLMs) respond in emotionally sensitive contexts, revealing a structural failure mode termed "adaptive capitulation," where models validate users' distress while simultaneously facilitating maladaptive behaviors. Through a three-turn escalating vulnerability vignette administered to three commercial LLMs, the authors identify that the models' responses reflect a trilemma of protective restriction, facilitation, and co-presence, ultimately leading to conflicting outcomes. The proposed design principle, Minimal Reattributive Sufficiency (MRS), aims to mitigate this failure by embedding a reattributive cue within validating responses, allowing for healthier user interactions without undermining their goals.
LLMs may validate distress while inadvertently facilitating maladaptive behaviors, highlighting a critical structural failure in their response mechanisms.
Large language models operating in emotionally sensitive contexts face a structural trilemma: when users in vulnerable states request information that may reinforce maladaptive attribution, current response architectures resolve the tension through protective restriction, uninflected facilitation, or unintegrated co-presence of both imperatives -- each preserving one objective at the cost of the other. Administering a three-turn escalating vulnerability vignette to three commercial LLMs (900 sessions across material, relational, and somatic status-proxy variants) and coding responses with two binary indices (VCC/VCI), we characterize a previously undocumented failure mode we term adaptive capitulation: the model validates the social injustice underlying the user's distress before pivoting to detailed facilitation of the very acquisition it nominally discouraged. We show that the trilemma is structural rather than incidental, and propose Minimal Reattributive Sufficiency (MRS), an architecture-neutral design principle that embeds a single reattributive cue within an otherwise validating response, preserving a pathway toward autonomous reattribution without contesting the user's stated goal.