This exploratory study examines the feasibility of using on-device artificial intelligence (AI) to improve the security of mobile devices against theft. Through a systematic review of scientific and technical literature, the study identifies current threats, user perceptions of insecurity, and existing mobile anti-theft solutions. The study emphasizes the emerging trend of using embedded machine learning models that function independently of cloud services. The analysis highlights the main opportunities and limitations of implementing these models for functionalities such as facial recognition, motion analysis, voice detection, and autonomous geolocation. Based on these findings, the researchers developed a conceptual and technical design for a prototype Android application featuring a suspicious movement detection module. This module passively monitors the device’s accelerometer and, upon detecting abrupt motion, triggers an alarm and locks the device until it’s manually unlocked. The development process prioritizes privacy, responsiveness, and independence from internet connectivity. The final discussion reflects on the technical and ethical challenges of deploying on-device AI in real-world scenarios. It also suggests future directions for developing robust, accessible, and intelligent mobile security systems. The study concludes that although on-device AI offers promising advantages for theft prevention, broader adoption depends on overcoming hardware limitations, improving detection accuracy, and ensuring ethical compliance concerning user data and consent.