Policy-Masked Private Experts: Auditable and Reversible Capability Access Control in Sparse MoE Models

arXiv:2608.06690 · cs.CR, cs.AI, cs.LG · Submitted 2026-08-11 · Read on arXiv

Zhuoheng Huang, Mukesh Singh

cs.CR, cs.AI, cs.LG

Submitted: 2026-08-11

Updated: 2026-08-12

Comments: 14 pages, 1 figure, 8 tables. Corrected the DeepSeek decoding and evaluation results and updated the corresponding analysis; corrected the mathematical statement of the frozen-parameter guarantee; clarified paired statistical inference; expanded related work and corrected references

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Benchmark — 25× in this paper · explained in 2 episodes
• Harness — 10× in this paper · explained in 2 episodes
• Transformer — 6× in this paper · explained in 2 episodes
• Trace — 4× in this paper · explained in 2 episodes
• KV cache — 3× in this paper · explained in 4 episodes
• Cross-Model Transfer — 2× in this paper · explained in 2 episodes
• Provenance — 2× in this paper · explained in 5 episodes
• Accuracy — 1× in this paper · explained in 1 episode
• Adam — 1× in this paper · explained in 3 episodes
• Calibration — 1× in this paper · explained in 9 episodes
• Clean Tool-Use Rate — 1× in this paper
• Continual Learning — 1× in this paper · explained in 3 episodes
• Decoder — 1× in this paper · explained in 2 episodes
• EMMA — 1× in this paper · explained in 2 episodes
• Embeddings — 1× in this paper · explained in 2 episodes
• Frozen model — 1× in this paper · explained in 3 episodes
• Generalization — 1× in this paper · explained in 14 episodes
• Localization — 1× in this paper · explained in 2 episodes
• Machine Unlearning — 1× in this paper · explained in 3 episodes
• Mixture-of-Experts (MoE) — 1× in this paper · explained in 11 episodes
• Papers — 1× in this paper · explained in 3 episodes
• Parameterization — 1× in this paper · explained in 2 episodes
• Sparsity — 1× in this paper · explained in 5 episodes
• Transformers — 1× in this paper · explained in 2 episodes
• Verification — 1× in this paper · explained in 2 episodes

Sources

• Who's Harry Potter? Approximate Unlearning in LLMs
• Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations
• Mixtral of Experts
• Gradient Routing: Masking Gradients to Localize Computation in Neural Networks
• DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
• The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions
• GateBreaker: Gate-Guided Attacks on Mixture-of-Expert LLMs
• RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs
• Qwen3 Technical Report
• ToolMind Technical Report: A Large-Scale, Reasoning-Enhanced Tool-Use Dataset
• ToolFailBench: Diagnosing Tool-Use Failures in LLM Agents
• CCTU: A Benchmark for Tool Use under Complex Constraints

Related papers

• SoK: AI-Augmented Binary Reversing
• Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
• Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
• Estimating the Social Cost of Corporate Data Breaches
• Efficient Fuzzy PSI under One-Sided Assumptions
• Sealing the Audit-Runtime Gap for LLM Skills