AI-assisted electrochemical sensors for drugs of abuse

The rapid development of new psychoactive substances (NPS) has intensified the demand for portable, cost-effective alternatives to conventional chromatographic methods. Electrochemical sensors offer advantages like rapid response, miniaturization potential, and low cost, yet their real-world deployment remains constrained by matrix effects, signal overlap, and the lack of genuine integration of artificial intelligence (AI) within the analytical workflow. This critical review examines recent advances (2024–2026) in machine learning (ML)-powered electrochemical sensors for the detection of drugs of abuse, assessing whether AI is solely used as post-acquisition data-analysis tool or embedded within the analytical workflow. Two detection strategies were identified: electrochemical fingerprinting on various platforms and biomimetic approaches employing aptamers or molecularly imprinted polymers (MIPs). Although ML algorithms can improve classification accuracy, they cannot completely resolve the issues of low selectivity and sensitivity of electrochemical sensors. Dataset quality, algorithm selection, applications in real samples, and model robustness were critically analyzed. Findability, accessible, interoperable, and reusable (FAIR)-compliant datasets, multi-analyte sensing platforms, and explainable AI (XAI) methods represent priorities for enabling reliable, decentralized drug monitoring in real-world scenarios. The most notable gaps are related to the absence of embedded on-device ML implementation, insufficient and non-standardized training datasets, inadequate assessment of real matrix variability, and lack of XAI approaches required for forensic and clinical validation.