Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper critiques the prevailing notion that local execution of AI models ensures user privacy, highlighting that local assistants can still aggregate sensitive data from various sources within an operating system. It introduces an OS-centered privacy framework that emphasizes institutional accountability and outlines a comprehensive threat model alongside a six-part privacy risk taxonomy. The findings reveal that effective privacy in on-device AI hinges on controlled information flow, limited authority, transparent user control, and robust governance mechanisms throughout the OS lifecycle.
Local execution of AI models doesn't guarantee privacy; a new framework reveals critical gaps in how on-device AI manages sensitive data across operating systems.
As AI systems move into operating systems, privacy no longer turns only on whether a model runs locally. A local assistant may assemble email, calendar entries, files, screenshots, notifications, and app intents; retain embeddings or summaries; invoke tools; emit telemetry; or route difficult requests to cloud infrastructure. Local inference reduces some exposure, but it answers only one question: where computation occurs. It does not answer who may assemble context, what derived state persists, which actions are authorized, or how updates change the system's authority. We develop an OS-centered privacy framework for on-device AI that treats privacy as an institutional accountability problem rather than a deployment attribute. The framework specifies a threat model, a six-part privacy risk taxonomy, privacy-by-architecture controls, and a four-level audit rubric. We demonstrate the rubric through a documentation-bounded comparison of Apple Intelligence/Foundation Models, Android AICore/Gemini Nano, and Microsoft Recall. Meaningful privacy in on-device AI depends on constrained information flow, bounded authority, visible user control, and auditable governance across the operating-system lifecycle.