AI-Driven Detection and Support for Hidden Addiction Patterns in Remote Workers: A Multimodal Approach

Abstract The widespread adoption of remote work has brought new flexibility to organizations but has also complicated the monitoring of employee well-being, particularly in identifying hidden addiction patterns such as substance abuse, behavioral compulsions, and digital overuse. This study introduces a privacy-preserving, AI-driven framework that leverages lightweight transformer models and multimodal behavioral analytics to detect and support addiction risks within distributed workforces. The proposed system integrates methodologies inspired by the mhGPT (Mental Health GPT) model and incorporates federated learning, differential privacy, and explainable AI to ensure both effectiveness and ethical compliance with GDPR and HIPAA standards. Model development and evaluation primarily utilized synthetically generated datasets and large-scale public datasets, with a limited-response survey informing feature design and scenario construction. Simulated experiments demonstrated high F1-scores in early risk detection and promising engagement rates for a tiered intervention protocol. While these results highlight the framework’s potential, real-world validation and further empirical study are needed to assess practical applicability and address ethical considerations in deployment. Keywords: Remote work, addiction detection, artificial intelligence, multimodal analytics, mental health, privacy