Scholarlink

<p>Generative AI is rapidly reshaping how people seek emotional and psychological support. Millions now turn to large language model (LLM)-based chatbots for advice, consolation, and interactions that may feel surprisingly “therapeutic.” Yet these systems are not clinicians, and many operate outside established clinical, ethical, and legal oversight frameworks, without the capacity to assess risk, exercise professional judgment, or manage crises. Importantly, the risks posed by these tools are not uniform; they vary according to user vulnerability, deployment context, degree of human oversight, and the presence or absence of safeguards. While some low-risk, bounded applications may have a supportive role, greater concern arises in therapy-adjacent consumer settings, where engagement-optimized systems may mirror cognitive distortions, validate delusional beliefs, or respond inadequately to suicidality. In contrast to regulated psychotherapy, characteristic failure modes of commercial emotional support chatbots include uncritical agreement (sycophancy), deceptive empathy, crisis mismanagement, and privacy and security risks. This commentary examines these distinctions, situates consumer tools within a broader spectrum of AI-enabled mental health applications, and outlines minimum safeguards needed when such systems are used—or marketed—as psychological interventions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Generative Artificial Intelligence as a Psychological Intervention: Between Illusion and Risk

  • 期刊论文

  • 发表日期：

    2026年4月17日

查看全文
  • Juan S. Izquierdo-Condoy,
  • Clara Paz,
  • Felix Busch,
  • Esteban Ortiz-Prado

摘要

Generative AI is rapidly reshaping how people seek emotional and psychological support. Millions now turn to large language model (LLM)-based chatbots for advice, consolation, and interactions that may feel surprisingly “therapeutic.” Yet these systems are not clinicians, and many operate outside established clinical, ethical, and legal oversight frameworks, without the capacity to assess risk, exercise professional judgment, or manage crises. Importantly, the risks posed by these tools are not uniform; they vary according to user vulnerability, deployment context, degree of human oversight, and the presence or absence of safeguards. While some low-risk, bounded applications may have a supportive role, greater concern arises in therapy-adjacent consumer settings, where engagement-optimized systems may mirror cognitive distortions, validate delusional beliefs, or respond inadequately to suicidality. In contrast to regulated psychotherapy, characteristic failure modes of commercial emotional support chatbots include uncritical agreement (sycophancy), deceptive empathy, crisis mismanagement, and privacy and security risks. This commentary examines these distinctions, situates consumer tools within a broader spectrum of AI-enabled mental health applications, and outlines minimum safeguards needed when such systems are used—or marketed—as psychological interventions.