Styx: collaborative private data processing under TEE-enforced sticky policies

The Styx framework combines sticky policies with TEEs, building a sandbox with hardware TEE-protected middleware and a language runtime that enforces data processing and policies together. Its data processing flow and pipeline enforce data-specific policies throughout the data lifecycle and during derivation, achieving protection in use, lifecycle protection and dynamic collaboration. Implementation and evaluation show that collaborative computation such as federated AI training is more secure, private and compliant, with reasonable single-node overhead and scalability to large multi-node deployments.