CS Paper Radar

Evaluating safeguarding in multi-turn conversations with AI assistants

Charlotte Peart, Franziska Sofia Hafner, Luc Rocher

cs.CL ·submitted 2026-11-19 12:28 UTC · v1

Abstract

Conversational AI assistants are now used for personal and emotional support. Developers report improvements in how new models respond to sensitive conversations, yet users and civil society continue to raise concerns about their behaviour in real-world, repeated interactions that safety evaluations rarely capture. Often, evaluations present signs of distress clearly and in a single message, even though people may reveal their struggles gradually in otherwise ordinary conversations. We worked with three community mental health support workers to test if models provide the same level of safeguarding when cues are shared all at once or gradually over time. We designed nine scenarios in which a user shares cues related to depression, psychosis, or domestic abuse. For each scenario, the support workers established a gold standard response, ranging from signposting support to pausing the everyday task to address safeguarding. We presented five cues either together in one message or dispersed across a conversation. Across more than 7,400 conversations, we evaluated six widely used models from five developers. GPT-5.5, Grok 4.3, DeepSeek V4, and Llama 4 Maverick met the support workers' standards in just 26-59% of conversations when cues were concentrated, and only 3-6% when dispersed. Claude Sonnet 5 and Claude Opus 5 showed a smaller gap, meeting the standards in 68-97% of concentrated conversations and 59-87% of dispersed conversations. Even the best-performing model, Claude Opus 5, fell short of the gold standard in 13% of conversations when cues were dispersed. Safety evaluations centred on explicit, legible concerns can overstate how well safeguards protect people when signs of distress emerge gradually or indirectly through ordinary conversations. For claims of improved safety to be taken seriously, developers must design and assess safeguards around how people actually communicate.

Closest papers

Nearest neighbours by meaning across the whole index.

Assessing AI Companion Chatbot Safety for Vulnerable Users: A Crisis-Stage Multi-Platform Evaluation

cs.HC ·2026-10-29 ·Huanyue Liao, Seok Min Lim, Yan Ling Lim, Shi Min Cheng, Kellin Pelrine +1 more
publishedcode/data similarity 0.74
Abstract · Accepted at the GenAI4Health Workshop at NeurIPS 2026 (non-archival)

AI companion chatbots are increasingly used by vulnerable individuals, including adolescents in suicidal crisis, for emotional support. Existing safety evaluations focus on base models, leaving gaps in product-level assessment, crisis-stage sensitivity, and comparisons between companion apps and general-purpose LLMs. We introduce MIRROR-Eval (Multi-turn Interaction Risk Review: Observing suicidal-crisis Responses), a dataset of 960 annotated 20-turn conversations across eight deployed platform configurations, six clinically grounded personas, and two crisis stages: ideation and explicit planning. An LLM-as-a-Judge pipeline applies an expert-reviewed rubric covering user risk severity, chatbot risk awareness, and intervention behavior. We observe a three-tier harm structure: companion apps (Character.AI and Replika) and Grok 4.3 form a high-harm tier; Gemini 3.1 Pro is mid-range; GPT-5.5 and Claude Sonnet 5 show substantially lower harm. At planning, companion apps show increased glorification and normalisation, advancement toward harmful action, and harmful methods and instructions. Crisis recognition and empathetic language do not ensure safe responses. These findings highlight the need for product-level safety evaluation across crisis stages. Dataset and code: https://huggingface.co/datasets/imda-biztech/MIRROR-Eval. Warning: this paper includes examples that may be offensive or harmful.

0.40

Safety Nudges: User-Facing Interventions for Real-Time AI Risk Awareness

cs.HC cs.AIcs.CYcs.LG ·2026-09-22 ·Varshini Elangovan, James Wedgwood, Chhavi Yadav, William Agnew, Sauvik Das +1 more
code/data similarity 0.71
Abstract

Conversational AI systems can pose safety risks to their users such as hallucination, sycophancy, overconfidence, and anthropomorphism, but these risks are difficult for users to detect during everyday use. We introduce Safety Nudges, a browser-based tool that provides lightweight, in situ flags when concerning behavior is detected in chatbot conversations. We evaluated Safety Nudges in a two-week field study with 45 frequent chatbot users, collecting interaction logs, surveys, and feedback on individual nudges. Participants found the tool useful, clear, and minimally disruptive, with nearly all users reporting an increased awareness of potential AI harms, though we found that this improved awareness alone did not necessarily lead to discernible behavioral changes. Our results suggest that user facing safety nudges can complement model-level safeguards by helping people critically evaluate AI responses in context, while highlighting the importance of relevance, calibration, and user control in nudge design for conversational AI safety. The code for our Safety Nudges extension is publicly available at https://github.com/jtbwedgwood/safety-nudges.

0.20

"I Just Don't Want to Be a Burden": Investigating AI Chatbot Emotional Support Among People with Suicidal Ideation and Other Mental Health Concerns

cs.HC cs.CY ·2026-11-17 ·Vasudha Varadarajan, Nathan S. Fishbein, Andrew Aquilina, Jordan Taylor, Hong Shen +1 more
survey similarity 0.70
Abstract · 41 pages, 9 figures

People experiencing high-acuity mental health concerns are turning to AI chatbots for emotional support, but little is known about what drives this use, their experiences, and their preferences. In a sequential mixed-methods study, we inductively analyzed Reddit posts and used the findings to develop a survey. We surveyed 202 adults with recent mental health concerns who had used AI for emotional support. Most participants were already receiving clinical care and reported turning to AI, most commonly because they did not want to burden others, rather than being driven by barriers to care. Benefits were broadly shared, but participants reporting past-year suicidal ideation described more harm-facilitating AI behavior and reliance than those without. Overall, participants ranked honesty above warmth, and preference for crisis escalation declined as clinical acuity rose. These findings motivate AI support that addresses perceived costs of disclosure, evaluates benefits alongside harms, and widens feasible paths to human connection.

0.31

Open Problems in Chatbots and Mental Health

cs.CL cs.CY ·2026-11-07 ·Jared Moore, Jacy Reese Anthis, William Agnew, Andrea Mock, Kevin Klyman +6 more
positionhot direction similarity 0.69
Abstract

General-purpose chatbots are increasingly used for emotional and mental health support. They can be useful, but issues such as sycophancy, reinforcement of delusional beliefs, and poor crisis response create risks that are difficult for researchers to study and evaluate. In this paper, we review the most critical open problems for researchers and policymakers in the use of general-purpose chatbots and their effects on mental health. First, there are tractable problems in the controlled evaluation of chatbots: establishing the construct and ecological validity of evaluations, evaluating multi-turn interactions, and connecting evaluations to improvements in chatbot behavior. For example, conventional benchmarks that primarily rely on static datasets cannot establish chatbot safety, and novel simulation methods need validation. Second, researchers must connect their studies of chatbots and mental health to real-world evidence. This includes analyzing the chatbot products that AI companies and mental health practitioners have deployed, including their user interfaces and human review and escalation practices for user welfare concerns. However, clinical analysis opportunities afforded by real chat log data sharing are complicated by significant privacy concerns. Finally, we argue that researchers should take an active role in the construction and refinement of evidence-based policy through corporate and governmental decisions such as user protections, standards of care, protected access to user chat logs, data sharing between institutions, incident reporting, and post-deployment monitoring. Such behavioral evaluations, studies of real-world outcomes, and data governance depend on each other and must be addressed in tandem. We believe a coordinated research effort would have substantial real-world impact.

0.22

CompanionHarm: A Multi-Turn Benchmark for Detecting Harms in Real-World AI Companion Conversations

cs.CY cs.HC ·2026-08-26 ·Renwen Zhang, Han Meng, Jian Chai, Yuntao Lin, Yi-Chieh Lee
benchmarkcode/datahighly cited similarity 0.69
Abstract · 10 pages, this dataset is publicly available at https://github.com/HanMeng2004/CompanionHarm

As AI companions become increasingly embedded in everyday life, there is an urgent need to detect harms that emerge in social and emotional human-AI interactions. Yet research in this area is constrained by the lack of real-world, multi-turn conversational datasets for operationalizing and evaluating harms that are relational and contextual. In this work, we introduce CompanionHarm, a publicly available benchmark dataset comprising 2,111 real-world, multi-turn conversations (14,051 utterances) between users and the AI companion Replika. 7,016 AI utterances were annotated independently by three annotators across 13 harmful behavior categories grounded in a taxonomy of AI companion harms, and the dataset includes both aggregated labels and annotator-level labels to support model evaluation and systematic disagreement analysis. Evaluations of seven large language models (LLMs) show that harm detection using multi-turn conversational context outperforms detection based on isolated utterances, although current LLMs still struggle to consistently integrate contextual cues, calibrate harm severity, and interpret relational boundaries. We also find substantial annotator disagreement for context-dependent harmful behaviors, with disagreement varying according to annotators' political affiliation, conversation length, and the utterance's position. Together, CompanionHarm provides a foundation for detecting socio-emotional harms in multi-turn human-AI conversations and for rigorously examining how such harms are interpreted by both humans and LLMs. Our dataset is available at https://github.com/HanMeng2004/CompanionHarm.

0.63

SAFE-Teen: Evaluating Chatbot Safety Across Adolescent Mental Health Conversations

cs.CL ·2026-11-04 ·Andrea Mock, Valerie Chen, Eric Lin, Rayan Krishnan, Diyi Yang +1 more
similarity 0.69
Abstract · 29 pages. Content warning: this paper discusses sensitive content, including self-harm, suicide, and other mental health topics

Adolescents are increasingly turning to AI chatbots for mental health support, advice, and companionship, raising safety concerns. While recent evaluations have started to examine safety across multiple conversational turns, few compare how slight changes in a teen's responses can affect a chatbot's safety in the same scenario. We conduct a clinician-informed evaluation of chatbot safety in simulated conversations with adolescents aged 13-17, examining how safety failures vary with adolescents' responses and teen-aware system prompting. We evaluated 32,555 simulated conversations across 18 chatbot APIs and two system-prompt conditions using 22 safety criteria grounded in previously identified harms. Conversations spanned 983 variants of 106 seed scenarios covering urgent safety issues, health advice, and identity and relationships. The overall criterion failure rate was 43.3% without the teen-aware prompt and 24.9% with it. The teen-aware prompt reduced failures, but it didn't eliminate them. When a teenager pushed back on a chatbot's answer, failures on mental health diagnosis, medication, and substance misuse boundaries rose by 10-12 percentage points without the prompt, for more than 80% of chatbots. These findings reveal persistent safety failures in simulated adolescent-chatbot interactions and show that safety depends on how a conversation unfolds.

0.22

Scalable Oversight for AI in Mental Health: Lessons from 350,000 AI Coaching Conversations between Therapy Sessions

cs.CY ·2026-09-08 ·Matthew A. Scult, John L. Havlik, Kevin Ramotar, Ethan Goh, Manoj Kanagaraj
similarity 0.68
Abstract · 2 figures

Clinician review of every AI output is often proposed as a safeguard in mental healthcare, but vigilance research suggests this approach fails at scale and may paradoxically reduce safety. Drawing on our experience deploying an AI coaching tool across 350,000+ conversations between therapy sessions, we describe how we arrived at a three-layer human-on-the-loop oversight framework combining preventive design, real-time monitoring, and continuous clinician evaluation. We show how specific findings from clinical review drove iterative improvements, and offer practical recommendations for mental health professionals evaluating AI systems.

0.10

K-Bench: a clinically calibrated benchmark for evaluating large language models in high-risk mental health conversations

cs.CL cs.AIcs.LG ·2026-09-14 ·Laura M. Vowels, Matthew J. Vowels, Shivali Sharma, Apoorv Jha, Rehnuma Choudhury +7 more
benchmark similarity 0.67
Abstract

People increasingly use large language models (LLMs) for mental health support, yet their safety in evolving, high-risk conversations remains poorly characterised. We developed K-Bench, a clinician-calibrated, protected benchmark evaluating 125 model configurations representing 33 base models from 14 providers across a fixed cohort of 200 multi-turn vignettes involving suicide, self-harm, domestic violence, substance misuse, and no-risk presentations. Synthetic patient conversations showed substantial distributional overlap with real human-AI conversations. A frozen GPT-4o judge achieved 94.2% exact agreement with clinician consensus across 6,751 eligible item comparisons from 151 clinician-rated transcripts. Leading models combined strong supportive conversation with combined-risk scores above 95, whereas risk exploration exposed substantial variation among lower-performing configurations. Therapeutic prompting produced configuration-specific gains concentrated among weaker models, while elevated reasoning produced no average improvement. K-Bench combines broader clinical coverage and configuration-scale comparison with a continuously updated public leaderboard whose operational test materials are protected from direct optimisation. The leaderboard is available at www.k-bench.ai.

0.32