Supplemental LLM Processing With Attested Encrypted Intermediary Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models, such as large language models (LLMs), pose significant concerns regarding user privacy as they may exploit user data for unintended purposes, lack transparency in data processing, and fail to prioritize user consent and accountability.
Innovation Solution
A system where client devices encrypt intermediary data using public-key attestations from server systems, employ anonymized tokens for access, and enforce immutable system properties through secure hardware to ensure secure and private communication, thereby preserving user privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LLM models process user data to improve accuracy and capabilities, then model performance is improved, but user privacy and data security deteriorate
Solution Approach 1:
The system segments the LLM processing into two parts: a local client-side model that processes user data privately, and a remote server that provides supplemental information. This segmentation allows the client to maintain privacy while still accessing enhanced capabilities through the server's additional data and models.
Solution Approach 2:
The patent introduces an intermediary encrypted communication channel between the client and server. The server acts as a mediator that provides supplemental information without directly accessing or exposing user data, as all communications are encrypted end-to-end using public-key cryptography.
2Adaptability or versatility
If server systems access user data to provide supplemental information, then model capabilities are enhanced, but data security and user consent control worsen
Solution Approach 1:
The system performs preliminary encryption of user data on the client side before transmission to the server. Public keys are generated and exchanged in advance, establishing secure communication channels before any data processing occurs. This preliminary security setup ensures that user data remains encrypted throughout the entire processing pipeline.
Solution Approach 2:
The patent implements local quality by keeping sensitive user data processing localized to the client device where it remains encrypted, while only non-sensitive supplemental information is processed on the remote server. Each system component performs only the specific functions it is designed for, with the client maintaining control over its own data.
3Object-affected harmful factors
If end-to-end encryption is implemented to protect user privacy, then user privacy is improved, but system complexity and key management burden increase
Solution Approach 1:
The system implements self-service by automatically generating and managing encryption keys on the client device without requiring manual user intervention. The client device autonomously handles key generation, storage, and usage, reducing the complexity burden on users while maintaining strong encryption protections.
Data Source
AI summary
Techniques are disclosed relating to improving user privacy using machine learning (ML) models. In various embodiments, a device processes a query using a locally stored large language model (LLM) operable to use supplemental data provided by one of a plurality of assisting server systems. The device verifies a set of public-key attestations, each attesting to a public key of a respective one of the assisting server systems. The device sends, based on the verifying, a request for the supplemental data to the assisting server systems. The request includes intermediary data produced by the processing and is encrypted using the attested-to public keys. The device processes the received supplemental data using the LLM to produce a result of the query. In some embodiments, the device encrypts the intermediary data with a symmetric key and encrypts a respective instance of the symmetric key with each of the attested-to public keys.