Private ML Model Hosting With TEE Key Sharing for Secure Collaboration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Enterprises lack the technical expertise to manage their own machine learning models and are hesitant to outsource them due to privacy concerns, especially when collaborating with untrusted service providers, as current Trusted Execution Environments (TEEs) do not adequately support secure collaborative computations among mutually untrusting parties.
Innovation Solution
A system and method utilizing Trusted Execution Environments (TEEs) in combination with cryptographic techniques, specifically Secure Multiparty Computation (SMPC), to securely distribute TEE management across multiple parties, enabling secure outsourcing and collaborative computations while ensuring data privacy through a 'ZeroTrust Secure Module' (ZSM) that requires participation from all parties for decryption and access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If enterprises outsource machine learning model management to service providers, then technical sophistication and ease of operation are improved, but security and privacy protection deteriorate due to potential hacking and data loss
Solution Approach 1:
The patent divides the TEE management control into separate shares held by different parties (service provider and customer). The service provider manages the TEE infrastructure while the customer holds cryptographic keys and data. This segmentation ensures that neither party has complete control, preventing single-point failures and unauthorized access while maintaining operational ease through distributed responsibility.
Solution Approach 2:
The patent introduces cryptographic key management as an intermediary mechanism between the service provider and customer. The key host system acts as a mediator that enforces access control policies, requiring mutual authentication and authorization before allowing any data access or computation. This intermediary layer maintains security while enabling convenient outsourcing of model management.
2Device complexity
If TEE management is centralized for ease of operation, then device complexity is reduced, but security deteriorates because a single compromised party can access all data and models
Solution Approach 1:
The patent implements segmentation of TEE management by distributing control authority between service provider and customer through separate key shares and access control shares. The service provider manages infrastructure operations while the customer controls cryptographic access. This segmentation prevents any single party from compromising the entire system, reducing the impact of harmful factors while maintaining manageable complexity through clear role separation.
Solution Approach 2:
The patent applies local quality by assigning different levels of access and control to different parties based on their specific roles. The service provider has local control over infrastructure management, while the customer has local control over data access and computation authorization. This differentiated access control ensures that each party can operate within their designated scope without gaining unauthorized access to sensitive information.
3Reliability
If multiple parties co-manage TEE for enhanced security, then privacy protection is improved, but device complexity and operational difficulty increase
Solution Approach 1:
The patent introduces key host and access control share systems as intermediaries that automate and simplify multi-party coordination. These intermediary systems handle the complex cryptographic operations, key management, and access control protocols automatically, reducing the operational burden on individual parties while maintaining strong privacy protection through distributed control.
Solution Approach 2:
The patent implements self-service mechanisms where the TEE system automatically enforces access control policies and manages cryptographic operations without requiring manual intervention from each party. The system self-regulates based on pre-established keys and policies, simplifying collaborative computation while maintaining security. Parties can authorize computations and data access through automated protocols rather than complex manual coordination.
4Reliability
If enterprises host models locally to protect privacy, then security is improved, but ease of operation deteriorates due to lack of technical sophistication
Solution Approach 1:
The patent introduces service providers as intermediaries that handle the technical complexity of local model hosting, infrastructure management, and system maintenance. The service provider operates the TEE infrastructure and provides technical expertise, while the customer simply needs to provide data and authorize computations. This intermediary approach maintains privacy protection through local hosting while dramatically improving ease of operation by eliminating the need for customers to possess advanced technical skills.
Solution Approach 2:
The patent implements self-service capabilities where the TEE system automatically manages cryptographic operations, access control, and computation execution without requiring customers to manually configure complex security settings. The system self-manages key management, automatically enforces access policies, and handles the technical intricacies of secure computation, allowing customers to easily host models locally without technical sophistication.
Data Source
AI summary
Systems and methods are disclosed for offering a secure model as a service. A system can be configured to implement, via a trusted execution environment comprising an enclave virtual machine and a customer key host, a distributed privacy policy in which a master decryption key is split-shared between the trusted execution environment the customer key host; perform a decryption of the master decryption key according to the distributed privacy policy to obtain a decrypted master key; and, based on the decrypted master key, perform, in the trusted execution environment, a private collaborative computation using one or more of customer data and a customer model. Secure multiparty computation can be used to perform the decryption of the master decryption key.