Artificial intelligence (AI), and especially generative AI, has quickly become one of the hottest topics in the modern age. Where digital footprints are ubiquitous and data has become a valuable commodity, AI has emerged as a focal point of innovation and concern. With AI now being integrated into operating systems (OS) promising enhanced efficiency, it is crucial to recognize that along with these advancements come new risks. AI models, which come in various forms, are essentially predictive programs trained to recognize patterns and generate responses. Engineers use vast datasets, ranging from private purchases to publicly available and web-scraped data, to train these models. Much like human learning, AI models retain patterns from their training to respond to queries without revisiting the original source material. The newest flavor of AI coming to fashion is the integration of an AI model into the OS of devices like phones or computers. Operating as an extension of the OS itself, these AI models primarily access local device data. They are “trained” before installation and continue learning from user interactions to tailor responses. This model is touted to be helpful with scheduling, drafting, searches, and other simple queries. When tasked with a query that is beyond its capacity, the embedded AI model will forward the request to the larger AI model in the cloud for a larger model to handle. Presumably, the larger model will process, return a response, and delete information about that request.

Full report : The Risks of an Operating System Integrated with Artificial Intelligence.