Privacy in AI systems

How personal and organizational data actually moves through an AI system (from the prompt a user types, through the model provider and its subprocessors, into vector stores, memory, logs, and traces) and the engineering controls that keep exposure bounded at each stop: data minimization, provider evaluation, secrets handling, sensitive-category treatment, end-to-end deletion, and honest user disclosure. Framed throughout as an engineering discipline, not legal advice, for builders responsible for what a system collects, retains, and exposes.