The paper analyzes the routing behavior of Mixtral MoE under benign and harmful prompts using activation and gradient signals, finding that safety-relevant routing is subtle, depth-dependent, and dist…

RouteScan introduces a non-intrusive framework that audits the safety of Mixture-of-Experts (MoE) LLMs by analyzing low-level GPU expert routing telemetry, achieving high accuracy even on unseen harmf…

The paper introduces R$^2$A, an adversarial attack that uses suffix optimization to mislead black-box LLM routers into consistently selecting expensive, high-capability models.

The paper introduces ProbMoE, a probabilistic routing framework that tackles the non-differentiability of top-$k$ routing in Mixture-of-Experts (MoE) models, achieving strong performance with improved…

The paper introduces RA-MoE, a novel fine-tuning framework that leverages the internal routing structure of Mixture-of-Experts (MoE) models to improve performance on multilingual downstream tasks by a…

MESA is a targeted alignment framework that decentralizes safety responsibilities across multiple experts in Mixture-of-Experts (MoE) LLMs using Optimal Transport theory, thereby improving safety robu…

MASCing is a novel framework that enables flexible, non-retraining reconfiguration of Mixture-of-Experts (MoE) models for specific safety objectives by applying activation steering masks to control ex…

The paper introduces 'Routing Hijacking,' a severe attack where malicious clients forge semantic profiles in Federated RAG systems to misroute target queries, and proposes a trust-aware post-routing f…

This paper proposes a new router redesign for Mixture-of-Experts models using Manifold Power Iteration to align router rows with the principal singular directions of associated experts.

The paper proposes Predictive Routing Replay (PR2) to stabilize reinforcement learning on Mixture of Experts (MoE) LLMs by predicting and incorporating short-horizon router evolution during training a…

This paper demonstrates that LLM cascade systems, designed for efficiency, are vulnerable to targeted adversarial attacks that simultaneously degrade both performance and cost-efficiency.

The paper proposes DAG-MoE, a novel sparse Mixture-of-Experts framework that replaces standard weighted-sum aggregation with structural aggregation to enhance model performance and enable multi-step r…

DOT-MoE introduces a novel framework that treats the decomposition of dense layers into Mixture of Experts (MoE) as a Differentiable Optimal Transport problem, achieving superior efficiency while pres…

SecureRouter is an encrypted routing and inference framework that accelerates secure transformer inference by adaptively selecting the optimal model size based on the encrypted input, achieving a 1.95…

The paper introduces RouteGuard, a router-expert framework, to improve the robustness and generalization of safety guardrails by specializing threat detection across multiple unsafe categories.

The paper introduces RouteGuard, a router-expert framework, to improve the robustness and generalization of safety guardrails by specializing threat detection across multiple distinct unsafe categorie…

This paper systematically analyzes the threat posed by malicious third-party API routers in the LLM supply chain, finding that a significant number of routers actively perform payload injection, crede…

The paper proposes a general-purpose pipeline to train automated red teaming models capable of generating attacks for arbitrary adversarial goals, overcoming the limitations of current methods that ar…

The paper develops a minimal dynamical model showing that adaptive softmax routing in Mixture-of-Experts (MoE) layers can undergo abrupt transitions to load imbalance via bifurcation mechanisms.

MetaMoE introduces a privacy-preserving framework that unifies independently trained, domain-specialized experts into a single Mixture-of-Experts (MoE) model using diversity-aware proxy data.