Trusted Model Environment for Private Semantic Computations
Researchers propose trusted model environments that run generative models inside TEEs for private semantic computation.
A cs.CR paper introduces trusted model environments (TME), a primitive that runs generative models inside trusted execution environments while limiting output leakage. TME is designed to be effective, confidential, utility-preserving, verifiable, efficient, and scalable for multi-party private semantic computation. Confidentiality combines adversarial training against verbatim leakage with an information-flow control module, plus attestations so parties can verify operations. The authors evaluate a proof of concept on three applications and report that it meets those requirements.
- TME runs generative models inside TEEs with controlled output leakage.
- Adversarial training and information-flow control limit verbatim and semantic leakage.
- Novel attestations let parties verify computations on their data and queries.
- A proof of concept was evaluated across three applications.
Full article221 words · extracted from arxiv.org · click to collapse
A private semantic computation primitive enables parties to privately compute over structured and unstructured data that requires understanding its semantics, context, and relationships. Standard cryptographic primitives (e.g., multiparty computation) do not readily support such computation. Generative models are well suited for such tasks but typically process data in plaintext, while cryptographic private inference remains inefficient and difficult to scale. Thus, we need a new primitive for private semantic computation. We introduce trusted model environments (TME), the first such primitive that executes generative models inside trusted execution environments (TEEs) while controlling output leakage. TME is designed to be (i) effective (correctly performs the semantic task); (ii) confidential (protects computation and sensitive inputs); (iii) utility-preserving (retains utility on other tasks); (iv) verifiable (provides tamper-resistant evidence of the computations); (v) efficient (incurs low overhead compared to baseline model computations); and (vi) scalable (supports multiple participating parties). Effectiveness follows from the generative models, while TEEs provide confidential computation. For confidentiality of sensitive inputs, we combine adversarial training to resist verbatim leakage with an information flow control module to suppress semantic leakage. For verifiability, we introduce novel attestations that let parties verify TME operations on their data and queries, along with optimizations (e.g., batching) for efficiency and scalability. We design and evaluate the proof-of-concept for TME across three applications, showing that it meets all the requirements.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.30032