The rapid advancement of Artificial Intelligence of Things (AIoT) in smart pharmaceutics promises personalized and adaptive drug delivery systems with real-time monitoring and precise dosing. However, these interconnected systems face critical challenges in data privacy, cybersecurity threats, and compliance with “right to be forgotten” regulations such as GDPR and DPDPA. Current AIoT frameworks lack efficient mechanisms to selectively remove sensitive patient data from trained models without full retraining, increasing vulnerability to privacy attacks and model bias. This paper proposes a Machine Unlearning-Enabled Cyber-Resilient AIoT Framework for privacy-preserving and adaptive smart pharmaceutics. The framework integrates edge AIoT sensors, a personalized dosing prediction model, an efficient machine unlearning module, and a distributed digital twin-based cybersecurity layer for real-time detection of insider and external threats. Simulation results on healthcare IoT datasets demonstrate that the proposed framework reduces privacy leakage by up to 92%, maintains over 95% model accuracy after unlearning, and lowers computational overhead by 40–60% compared to complete retraining baselines. It also exhibits strong resilience against common AIoT cyberattacks while ensuring regulatory compliance. This work bridges pharmacy, artificial intelligence, and cybersecurity, offering a scalable privacy-by-design solution for next-generation personalized drug delivery systems.