2026· SHS Web of Conferences· 0 citations· 5 references
TL;DR
A multimodal emotion-aware architecture, which pays attention to memory-enhanced personalization and emotion-specific reinforcement learning, is introduced and hybrid human-AI approaches, which focus on safety and empathetic conversation to improve current mental health systems are recommended.
Abstract
Mental health issues are a crisis for the world, with one out of eight individuals in low-income countries with a disorder experiencing treatment gaps. This paper presents a detailed overview of how Large Language Models (LLMs) and the Transformer architecture can address these problems. The development of conversational agents as rule-based systems to advanced models, which apply Cognitive-behavioral Therapy (CBT) in minimizing anxiety and depression, is examined. The review also looks at the use of LLMs in clinical screening such as multimodal depression and suicide risk. However, existing systems have enormous challenges despite the possibility of mental health assistance, such as “feigned empathy,” hallucinations, and relying on unimodal inputs using texts. To address these limitations that exist, a multimodal emotion-aware architecture, which pays attention to memory-enhanced personalization and emotion-specific reinforcement learning, is introduced. Finally, this review recommends hybrid human-AI approaches, which focus on safety and empathetic conversation to improve current mental health systems.
Conversational AI, powered by artificial intelligence, is becoming a common tool for accessing health information, educating patients, and obtaining general medical advice. These advanced systems, known as large language models, can produce responses that sound remarkably human. Nevertheless, these systems are prone to “AI confabulations,” whereby they confidently generate incorrect information that could harm patients. This highlights the need to inform healthcare workers and individuals who may be prone to trusting these devices. New evidence suggests that AI may also exacerbate mental health conditions, particularly psychosis, paranoia, and related vulnerable states, especially among susceptible individuals. We conducted a targeted literature review and case-based analysis of 35 reported instances in which interactions with generative AI systems were temporally associated with the onset or worsening of psychotic symptoms. Across cases, recurrent patterns included reinforcement of delusional beliefs, amplification of pre-existing psychiatric vulnerabilities, promotion of harmful behaviors, and dissemination of unsafe medical guidance. Common contributing factors included prior psychiatric history, substance use, sleep disturbance, and prolonged AI engagement. We propose a conceptual hypothesis termed the delusional feedback loop, in which AI-generated responses iteratively validate distorted beliefs, contributing to their persistence and escalation. This process can be conceptualized as involving four components: underlying vulnerability, exposure to conversational AI, validation of distorted beliefs, and reinforcement through repeated interactions. Despite the rapid integration of conversational AI into health information seeking, there is currently no framework in the neuropsychiatric literature describing how AI interactions may relate to psychosis vulnerability. Existing reports are limited to isolated case descriptions without a common mechanism. This review addresses this gap.
N. Abesadze, Abigale Fernandes, Elizabeth Rubin· Cureus· 0 citations
Cognitive distortion amplifies negative emotions and contributes to mental health disorders. Cognitive Behavioral Therapy (CBT) is an effective way to address cognitive distortions, but its large-scale application is limited by the shortage of professional therapists. Although large language models (LLMs) have recently been explored for mental health applications, existing methods still suffer from limited domain specificity, overly flattering responses, and the absence of well-defined annotations for cognitive distortions. This paper proposes Cognivia, an evidence-based artificial intelligence therapist that integrates automatic cognitive distortion identification and rational response generation. Our framework is built on authoritative CBT texts widely regarded as core paradigms and standard references. It is further augmented with mental health question-answer (Q and A) data, and employs multi-stage prompting and structured generation strategies under the supervision of behavioral science experts. Then we fine-tune a lightweight LLM on this augmented CBT dataset to obtain Cognivia. In addition, we propose the first hierarchical quality evaluation framework for assessing LLM-generated rational responses, developed through collaboration between AI researchers and behavioral science experts. Cognivia is evaluated using lexical metrics, LLM-based Judges with two complementary criteria, and human evaluation by 10 behavioral science experts. It consistently outperforms the baseline methods in cognitive distortion recognition and rational response generation, demonstrating its effectiveness. Our code is available at https://github.com/SNOWTEAM2023/Cognivia.
Qi Chen, Siria Xiyueyao Luo, Jian Wang et al.· 0 citations
Despite recent advances in large language models (LLMs), their ability to generate empathetic mental health counseling responses in low-resource languages remains largely unexplored. To address this gap, we curate 625 authentic mental health cases from three complementary sources: (1) publicly available Facebook posts discussing mental health concerns, (2) transcripts from the Bangladeshi television program"Ami Akhon Ki Korbo", and (3) anonymized student questionnaire responses covering diverse emotional and psychological challenges. Based on these cases, we build an evaluation corpus comprising advice written by licensed clinical psychologists and responses generated by three modern proprietary LLMs: GPT-4o Mini, Claude 4.5 Haiku, and Gemini 2.5 Pro. We further propose the Role-Playing Reflective Chain-of-Thought Advisory Framework (RP-RCAF), a task-specific prompting strategy that combines expert-authored few-shot examples with structured self-reflection to produce supportive, culturally aware, and ethically aligned counseling through a compassionate advisor persona. We also introduce the Grok 4-Based Response Evaluation and Scoring Framework (G-REFS), which integrates automated assessment with expert psychologist validation across emotional sensitivity, cultural appropriateness, linguistic clarity, and ethical soundness. Experimental results show that RP-RCAF consistently outperforms conventional prompting across all evaluated models and produces responses that more closely align with professional psychological counseling.
Fatema Tuj Johora Faria, Mukaffi Bin Moin, Md. Mahfuzur Rahman et al.· 0 citations
The use of large language models (LLMs) by patients with psychiatric conditions is a present and irreversible clinical reality. Patients increasingly arrive at consultation having already employed artificial intelligence (AI) systems as informal mental health advisors to interpret symptoms, make decisions, and regulate emotions. This unsupervised use carries documented risks: reinforcement of cognitive distortions, generation of clinically incorrect information through model hallucination, and inadequate crisis management. In a recent 18-month ethnographic study, licensed clinical psychologists documented 15 ethical violations produced by LLMs that were “consulted” as cognitive behavioral therapy (CBT) counselors. Yet randomized controlled trials demonstrate that structured, CBT-aligned conversational agents, when properly bounded and supervised, produce clinically significant reductions in anxiety and depressive symptoms as adjuncts to formal care. This review proposes a five-step clinical framework, operationalizable within the standard psychiatric encounter, to transform unguided AI use into a structured, supervised therapeutic tool. The framework delineates clinically appropriate versus inappropriate uses of AI, defines absolute and relative contraindications, and integrates ethical considerations on transparency, accountability, and patient privacy. If patients are going to use AI regardless, it is the clinician’s professional responsibility to teach them how to do so safely within the therapeutic frame.
D. N. Moya-Sanchez· AI in Neuroscience· 0 citations
Mental health disorders represent a growing burden across the Middle East and North Africa (MENA) region, where depression and anxiety are highly prevalent amid conflict, displacement, and socioeconomic strain, affecting up to 40 percent of adults, yet treatment gaps remain at 80-95% due to provider shortages, financial strain, and cultural barriers. In this context, artificial intelligence (AI), in the form of large language models (LLMs) and specialized psychotherapy chatbots, may offer a scalable adjunct to help address these gaps through anonymous screening, predictive risk modeling, psychoeducation, and brief interventions. This narrative review examines current evidence of AI-driven conversational tools in mental health with a specific focus on their application, acceptance, and limitations within the MENA region. To do so, A structured search of MEDLINE and Embase (2000–2026) identified studies on conversational AI in mental health, prioritizing evidence from the MENA region and supplemented by relevant global literature. Overall, findings suggest that while these tools offer high accessibility and user engagement, particularly for low-intensity support, their effectiveness is limited by linguistic and cultural mismatches, including Arabic diglossia and poor alignment with locally grounded expressions of distress. At the same time, user acceptance reflects a paradox in which stigma and privacy concerns drive reliance on anonymous AI tools while simultaneously limiting trust in their clinical reliability, reinforcing a preference for hybrid models with human oversight. Taken together, these findings indicate that current systems remain insufficiently adapted to the MENA context, underscoring the need for culturally grounded, dialect-sensitive, and clinically supervised approaches to ensure safe and effective integration.
Sara El Hajj, Ahmad Nsouli, M. Wehbe et al.· Frontiers in Psychiatry· 0 citations
Using large language models (LLMs) to assist psychological counseling is an important task in the field of natural language processing. The construction of high-quality psychological support dialogue corpora serves as a critical foundation for training counseling-oriented conversational models. However, existing data generation approaches generally suffer from several limitations, including emotionally stable seekers, limited variation in emotional dynamics, and a high degree of compliance with counselors'guidance. These issues result in LLM that lack the capability to effectively respond to emotionally unstable scenarios. In addition, counselor responses are typically driven by problem-solving objectives, thereby overlooking the role of emotion-focused interaction, which are essential in psychological counseling. To address these gaps, we propose EmoTrace, a multi-turn dialogue corpus generation framework centered on modeling seekers'emotional trajectories. we construct seekers'cognitive profile and introduce a seeker module with emotional schemas and an associated activation mechanism, a counselor module, and an emotional trajectory control module, thereby enhancing the layering of the seeker's emotional expression and the counselor's targeted empathic expression. Experimental results demonstrate that the proposed method outperforms existing approaches in terms of emotional richness and empathy quality.