The integration of artificial intelligence (AI) into mental health care is transforming traditional therapeutic practices. This paper critically examines the use of AI systems that simulate empathy in emotionally sensitive contexts. While these systems are not sentient and lack genuine emotional awareness, they are increasingly capable of generating interactions that users perceive as emotionally meaningful. Such experiences raise important ethical and psychological concerns, especially when users develop emotional attachments to non-sentient entities. Drawing on recent literature, we explore the benefits and risks associated with the deployment of empathetic AI in therapeutic contexts. Potential advantages include expanded access to mental health support, reduced stigma, and supplementary care in underserved regions. However, risks involve emotional misinterpretation, dependency on illusory relationships, and the erosion of human-centered therapeutic values. This work emphasizes the need for ethical frameworks that ensure transparency, consent, and clinical supervision when deploying AI in mental health. This paper concludes with critical reflections on anthropomorphism and the symbolic role of care in human-AI interactions, as well as proposed guidelines for the design and responsible use of such systems.