Remembering Is Not Enough: Evaluating Timing and Omission in Longitudinal AI Mental Health Support

arXiv:2611.08721v1 Announce Type: new Abstract: Evaluations of large language models (LLMs) for mental health support commonly focus on the clinical appropriateness of individual responses, potentially overlooking failures that emerge across extended interactions. As AI assistants accumulate memory of users' circumstances and prior disclosures, this raises an accountability question: Do AI chatbots provide appropriate support as users' needs evolve? We investigate by analyzing 14,256 longitudinal simulations grounded in donated chatbot histories from 22 users, examining how a memory-equipped assistant responds to suicidal ideation, self-harm, and anxiety crises across recovery trajectories and communication styles. We introduce the Context-Aware and Readiness-Evaluative (CARE) framework, developed with mental health professionals, to assess the timing and quality of advice, including omissions. Advice was less aligned with users' readiness after a crisis than in matched control stretches of the same conversation, largely because of premature recommendations, which were most common when recovery stalled. Revisiting a crisis reduced premature advice but not lagging advice, and users' communication style changed how often and when the assistant gave recommendations. Although explicitly unproductive advice was rare, the assistant frequently omitted clinician-expected support such as follow-up and planning for next steps. These findings show the limits of single response-level safety evaluation and the need to assess timing, omission, and consistency across users when holding AI

AI chatbots may provide inappropriate support for users' evolving mental health needs, potentially causing harm.

The research highlights deficiencies in AI mental health support, such as premature and omitted advice, which could lead to inadequate assistance during critical times. However, it also acknowledges that explicitly unproductive advice was rare, suggesting some potential utility. The overall societal impact is mixed due to both risks and the importance of further refining AI accountability in longitudinal mental health contexts.

This is rigorous evidence that AI mental health chatbots, increasingly used for real support, mistime advice and omit clinically expected follow-up during crisis recovery. The implication is concerning given real-world deployment of such tools for vulnerable users, though the study itself is a careful diagnostic rather than a harm report.

The study reveals significant limitations in AI mental health support, specifically concerning inappropriate timing of advice and critical omissions in longitudinal care. These shortcomings, especially in sensitive areas like suicidal ideation, indicate a material risk to patient well-being. The findings highlight that current evaluation methods are inadequate for assessing real-world performance, underscoring a concerning gap in AI's readiness for such critical applications.

The study finds that AI mental health assistants often give premature or lagging advice and omit expected support. This indicates potential harm in real-world use, as users may not receive timely or appropriate care.

SOURCE     arXiv cs.CY (Computers and Society) (tier 1)
   ↓
DOCUMENT   33849db9-0677-4aae-b448-22f651191005
           https://arxiv.org/abs/2611.08721
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  5 runs, methodology 1.0.1
   ↓
SCORE      -61  (Adverse)
   ↓
CONFIDENCE 81%