Yuhan Ma, Stefan Schmid
Conference on Neural Information Processing Systems (NeurIPS) 2026 Accepted
A security architecture for tool-using LLM agents that confines sensitive plaintext at the read boundary and authorizes external actions through a trusted executor.
Yuhan Ma, Stefan Schmid
Conference on Neural Information Processing Systems (NeurIPS) 2026 Accepted
A security architecture for tool-using LLM agents that confines sensitive plaintext at the read boundary and authorizes external actions through a trusted executor.
Yuhan Ma, Yong Li, Stefan Schmid
International Conference on Machine Learning (ICML) 2026 Accepted
A function-secret-sharing based framework for efficient two-server secure LLM inference, targeting fixed-point comparisons, interval lookup, low-degree polynomial evaluation, and predicate-dependent computation.
Yuhan Ma, Yong Li, Stefan Schmid
International Conference on Machine Learning (ICML) 2026 Accepted
A function-secret-sharing based framework for efficient two-server secure LLM inference, targeting fixed-point comparisons, interval lookup, low-degree polynomial evaluation, and predicate-dependent computation.