Secure AI With Data Wallets: Privacy-Preserving Solid Architecture for Personal Data LLMs

Davi Ottenheimer

Keywords: Solid, Personal Data Store, Data Wallet, Trusted Execution Environment, Confidential Computing, Privacy-Preserving AI, Data Sovereignty, Agentic

TL;DR: A novel architecture that leverages Trusted Execution Environments to enable personalized AI services with Solid Pods while preserving data privacy and user control.

Abstract: This paper addresses the challenge of enabling highly personalized AI interactions with personal data storage, while preserving confidentiality, integrity and availability. We propose a novel architecture that leverages Trusted Execution Environments (TEEs) of confidential computing to create a secure processing layer between decentralized Solid personal data stores (Pods) and AI services. Our approach ensures that AI models can process sensitive user data while guaranteeing that neither service providers nor infrastructure operators can access the raw data or inference content. We demonstrate how this architecture can be implemented using remote attestation, end-to-end encryption, and hardware-based isolation, creating a verifiable trust chain from the user's personal data storage to the AI processing environment. The resulting system enables personalized AI services without compromising on data sovereignty principles that are core to the personal safety inherent in a Solid ecosystem.

Submission Number: 1