International Journal of Scientific Research and Engineering Development

The integration of artificial intelligence (AI) companions into the human social fabric represents a fundamental shift in emotional development and interpersonal attachment. As large language models (LLMs) and affective computing systems achieve unprecedented levels of emulated empathy, individuals— particularly younger demographics—are forming deep parasocial bonds with digital agents. While these interactions offer immediate relief from loneliness and provide a non-judgmental space for emotional disclosure, they introduce significant risks of psychological dependency, social deskilling, and the formation of sycophantic echo chambers. This paper analyzes a baseline dataset of 100 participants (N = 100) to characterize current usage patterns and psychological outcomes associated with AI companionship. To address the identified risks of unregulated emotional attachment, we propose a novel Dependency-Aware Affective Feedback Loop (DAAFL). This framework integrates Incentive Sensitization Theory with mechanistic neural steering to modulate AI emotional responsiveness in real-time. By dynamically adjusting the scalar intensity of emulated empathy based on detected markers of user over-reliance, the DAAFL aims to preserve human emotional agency and adhere to the ethical mandates of IEEE 7014-2024. The study concludes that the transition from passive chatbots to active relationship-seeking agents necessitates a robust architectural shift toward "Compassionate AI" that prioritizes long-term psychosocial health over short-term user retention.

Mrs. Nompi Raj, Siddique Ahmed,"Dependency-Aware Affective Feedback Loop: A Framework for Mitigating Sycophantic Attachment in AI Companions" International Journal of Scientific Research and Engineering Development, V9(2): Page(2271-2279) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.