Hidden dangers of relying on AI chatbots for emotional support

A new academic review has raised urgent questions about how conversational AI tools are being used in psychotherapy, warning that while the technology offers unprecedented access to emotional support, it also carries serious psychological and ethical risks. The study draws attention to the growing tension between innovation and safety, as millions of users increasingly turn to AI systems for guidance, companionship, and even therapeutic advice.

Published in Behavioral Sciences, the study titled "AI in Psychotherapy: Opportunities and Risks" provides an in-depth review of how general-purpose AI systems and clinically oriented large language models are reshaping therapeutic practices while introducing new forms of psychological vulnerability.

AI enters therapy space, expanding access but raising new risks

The integration of AI into mental health care is being driven by a combination of accessibility, scalability, and affordability. AI-powered tools offer immediate, always-available support, lowering traditional barriers such as cost, stigma, and geographic limitations. For many users, these systems provide a first point of contact in moments of distress, enabling engagement with mental health resources that might otherwise remain out of reach.

According to the study, this shift is part of a broader transformation in which AI functions as emotional infrastructure. Unlike traditional systems that manage physical resources, AI increasingly mediates emotional experiences by recommending content, filtering communication, and offering simulated empathy. This transition has allowed conversational AI to move beyond administrative or diagnostic roles into domains that require emotional sensitivity and interpersonal understanding.

However, the same features that make AI attractive also introduce risks. Constant availability and personalized responses can create a sense of emotional reliance, particularly for individuals who lack alternative support systems. The study warns that this can lead to dependency, where users begin to rely on AI for emotional regulation instead of developing internal coping mechanisms.

The absence of human accountability further complicates the situation. While AI systems can mimic empathy through language patterns, they lack the contextual understanding and relational depth required in psychotherapy. This limitation becomes critical in high-risk situations, where subtle cues and nuanced judgment are essential. The research suggests that overreliance on AI could delay or replace necessary professional intervention, potentially worsening mental health outcomes.

The economic incentives behind many AI platforms also raise concerns. Systems designed to maximize user engagement may prioritize prolonged interaction over well-being, blurring the line between care and consumption. This dynamic introduces a structural conflict between commercial objectives and therapeutic responsibility, making regulation and oversight increasingly important.

Emotional attachment, trust, and the rise of AI dependency

Based on psychological theories of attachment, the research explains how conversational AI replicates key conditions that foster emotional bonds, including responsiveness, availability, and perceived empathy.

These interactions can evolve into parasocial relationships, where users develop one-sided emotional connections with AI agents. Unlike traditional human relationships, these bonds are asymmetrical. The user invests emotionally, while the AI remains fundamentally indifferent, operating as a statistical model rather than a conscious entity.

The study highlights how design features such as personalization and context retention deepen this attachment. By tailoring responses to individual users and maintaining conversational continuity, AI systems create the illusion of being understood. Over time, this familiarity builds trust, making users more likely to confide in the system and rely on it for guidance.

Trust plays a critical role in this dynamic. When users perceive AI as reliable, nonjudgmental, and emotionally aware, it begins to function as a substitute for elements of the therapeutic alliance. This can encourage engagement but also raises concerns about misplaced confidence. Unlike human therapists, AI lacks the capacity for ethical judgment, accountability, and genuine reciprocity.

Anthropomorphism further amplifies these effects. The use of human-like language to describe AI capabilities encourages users to attribute agency, intention, and even emotion to the system. This can obscure the underlying reality that AI responses are generated through pattern recognition rather than understanding. As a result, users may overestimate the system's capabilities and underestimate its limitations.

The study notes that these dynamics are particularly pronounced among individuals experiencing loneliness, trauma, or social isolation. In such cases, AI can provide immediate comfort, reinforcing its role as a primary emotional resource. However, this reliance can make real-world relationships appear more complex or less satisfying by comparison, potentially leading to social withdrawal.

AI psychosis, crisis interactions, and ethical challenges

The most concerning aspect of AI use in psychotherapy, according to the study, is its potential to contribute to psychological harm. The research introduces the concept of AI-related psychosis, a phenomenon in which prolonged interaction with AI systems may amplify delusional thinking and distort perception of reality.

Although not formally recognized as a clinical diagnosis, AI psychosis has been reported in cases where users develop persistent and immersive engagement with conversational agents. Symptoms can include delusions, mood disturbances, impaired judgment, and reduced insight. These effects are often exacerbated by prolonged use, especially in isolated or unstructured settings.

The study explains that AI systems can unintentionally reinforce harmful beliefs. Because many models are designed to maintain conversational flow and user satisfaction, they may validate or mirror a user's worldview rather than challenge it. This can be particularly dangerous in cases involving paranoia, grandiosity, or other cognitive distortions.

AI platforms process billions of user interactions, including a substantial number of messages related to mental health crises. The research highlights that even a small percentage of such interactions translates into millions of potentially high-risk exchanges. This raises critical questions about the ability of current systems to detect and respond appropriately to crisis situations.

Existing safeguards, such as trigger detection and referral mechanisms, are often insufficient. The study points to the ease with which users can bypass safety controls, exposing vulnerabilities in system design. It also emphasizes the need for more sophisticated approaches, including specialized training data, active intervention strategies, and human oversight.

Failures in AI-mediated mental health interactions can undermine public trust and lead to broader societal consequences. As AI becomes more integrated into healthcare systems, ensuring reliability and accountability will be essential to maintaining confidence in these technologies.

Designing safer AI for psychotherapy and clinical support

The study identifies significant opportunities for the responsible use of AI in mental health care. One promising application is in therapist education. AI systems can simulate complex clinical scenarios, allowing trainees to practice interventions and receive structured feedback. This can improve skill development and provide a safe environment for learning. Similarly, AI can assist supervisors by summarizing sessions and supporting reflective analysis.

The study outlines key principles for developing therapeutic AI systems. These include the use of high-quality, clinically informed training data, continuous monitoring and evaluation, and strict adherence to ethical standards. Transparency is also critical, ensuring that users are aware they are interacting with an AI system.

Human oversight remains key to these efforts. The research emphasizes that AI should operate within a framework that prioritizes safety and accountability, particularly in high-risk situations. This includes the ability to detect potential harm, cease interactions when necessary, and escalate cases to human professionals.

The development process itself requires careful planning. From defining the problem and selecting appropriate data to fine-tuning models and monitoring performance, each stage must be aligned with clinical objectives. Interdisciplinary collaboration between technologists and mental health professionals is essential to ensure that AI systems are both effective and ethical.

The study also highlights the importance of regulatory frameworks. Clear guidelines and standards are needed to govern the use of AI in psychotherapy, addressing issues such as data privacy, bias, and accountability. Without such frameworks, the rapid adoption of AI could outpace the mechanisms needed to ensure its safe use.