RARE: Decoupling Representation Steering from Expert Routing in Mixture-of-Experts Language Models

• URL: http://arxiv.org/abs/2608.21236v1
• Date: Fri, 21 Aug 2026 15:45:36 GMT
• Title: RARE: Decoupling Representation Steering from Expert Routing in Mixture-of-Experts Language Models
• Abstract summary: We introduce a router-agnostic representation engineering framework for MoE language models.<n>RARE projects arbitrary behavioral perturbations onto the null space of the router matrix, thereby removing router-visible components.<n>On harmfulness steering, RARE reaches an average attack success rate of 53.3% while retaining 67.8% MMLU accuracy.
• Score: 13.298401993417611
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Representation engineering offers a lightweight means of controlling language-model behavior by modifying intermediate hidden states, but its direct application to Mixture-of-Experts (MoE) models introduces a structural mismatch. We first verify this failure mode through a series of empirical studies and find that preserving clean routing substantially recovers steering performance and that routing is more sensitive to semantic content than to behavioral changes under controlled content. Motivated by these findings, we introduce RARE, a router-agnostic representation engineering framework for MoE language models. RARE projects arbitrary behavioral perturbations onto the null space of the router matrix, thereby removing router-visible components, and further corrects routing drift propagated to selected downstream layers. To decide the best perturbation estimator in this framework, we evaluate five estimators on six heterogeneous open-weight MoE models across three steering scenarios: harmfulness, truthfulness, and factual editing. On harmfulness steering, RARE reaches an average attack success rate of 53.3% while retaining 67.8% MMLU accuracy, yielding a stronger aggregate effectiveness--utility trade-off than baselines. It further improves average TruthfulQA MC1 accuracy from 41.0% to 58.6% and CounterFact efficacy from 16.8% to 96.3%. These results support routing consistency as an important architectural consideration for adapting representation engineering to MoE models.

Related papers

• AutoCRAT: Within-trajectory Joint Control of Stochasticity and Compute for LLM Reasoning [10.408251801653723]
  Large language models (LLMs) achieve strong reasoning performance, which depends critically on time decisions.<n>Recent adaptive methods partially address this limitation, but they primarily adapt either recommended decodingity (how the model explores) or reasoning compute (how long the model reasons) in isolation.<n>We introduce a within-trajectory joint control view, and instantiate it in AutoCRAT.
  arXiv  Detail & Related papers  (2026-08-30T19:24:58Z)
• Grounding latent algorithm routing in transformer reasoning [2.3977651066739374]
  We study whether transformers can organize episode-level adaptation around different inductive-bias families.<n>We introduce ROUTEBENCH, a diagnostic benchmark whose regimes favor global shrinkage, sparsity, robustness, and locality.<n>Across dense decoder-only transformers trained from scratch at 44M-612M parameters, a 306M model closes 80.9 percent of the oracle-routing gap and achieves route F1 of 84.1.
  arXiv  Detail & Related papers  (2026-07-27T14:07:12Z)
• Towards Robust Training in NNGPT AutoML Pipeline: A Loss-Optimizer Pairing Selection Study [48.83701310501069]
  This paper investigates whether a single recipe is sufficient for heterogeneous architecture pools, or whether the optimal pairing varies across structurally diverse models.<n>We conduct a systematic empirical study of all $3 times 6 = 18$ combinations of six classifiers paired with three loss functions: Cross-Entropy (CEL), Negative Log-Likelihood (NLL), and the recently introduced genetically evolved NGL loss across the base models presented in LEMUR heterogeneous architecture pool on six image classification datasets.<n>Our results confirm that no single pairing is universally optimal. Cross-Entropy with Adam or AdamW is the most robust choice across
  arXiv  Detail & Related papers  (2026-06-18T20:51:42Z)
• RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs [9.24144060095413]
  We present RouteHijack, a routing-aware jailbreak for large language models (LLMs)<n> RouteHijack performs response-driven expert localization to identify safety-critical and harmful experts.<n>It then constructs adversarial suffixes with a routing-aware objective that suppresses safety experts, promotes harmful experts, and prevents early-stage refusal.
  arXiv  Detail & Related papers  (2026-05-01T11:54:55Z)
• MASCing: Configurable Mixture-of-Experts Behavior via Activation Steering Masks [12.70638634576904]
  We present MASCing, a framework that enables flexible reconfiguration of MoE behavior across diverse safety scenarios without retraining.<n> MASCing uses an LSTM-based surrogate model to capture cross-layer routing dependencies and map routing logits to downstream behaviors.<n>It then optimize a steering matrix to identify behavior-relevant expert circuits and apply steering masks to the routing gates to override expert selection.
  arXiv  Detail & Related papers  (2026-04-30T12:58:57Z)
• OOD-MMSafe: Advancing MLLM Safety from Harmful Intent to Hidden Consequences [64.01706941950489]
  Current safety paradigms primarily target malicious intent or situational violations.<n>We propose shifting the safety frontier toward consequence-driven safety, a paradigm essential for the robust deployment of autonomous and embodied agents.<n>We develop the Consequence-Aware Safety Policy Optimization (CASPO) framework, which integrates the model's intrinsic reasoning as a dynamic reference for token-level self-distillation rewards.
  arXiv  Detail & Related papers  (2026-03-10T14:16:43Z)
• ACAR: Adaptive Complexity Routing for Multi-Model Ensembles with Auditable Decision Traces [3.151184728006369]
  We present ACAR, a measurement framework for studying multi-model orchestration under auditable conditions.<n>ACAR uses self-consistency variance (sigma) computed from N=3 probe samples to route tasks across single-model, two-model, and three-model execution modes.<n>We evaluate ACAR on 1,510 tasks spanning four benchmarks, producing more than 7,550 auditable runs.
  arXiv  Detail & Related papers  (2026-02-06T23:27:17Z)
• Internalizing LLM Reasoning via Discovery and Replay of Latent Actions [4.830503861275364]
  Internalization of chain-of-thought processes into hidden states has emerged as a highly efficient paradigm for scaling test-time compute.<n>We propose STIR (Self-Distilled Tools for Internal Reasoning), a framework that reformulates reasoning enhancement as a dynamic latent trajectory control problem.
  arXiv  Detail & Related papers  (2026-02-04T08:44:57Z)
• RISER: Orchestrating Latent Reasoning Skills for Adaptive Activation Steering [62.63376387138257]
  We propose a plug-and-play intervention framework that adaptively steers large language models (LLMs) reasoning in activation space.<n>RISER constructs a library of reusable reasoning vectors and employs a lightweight Router to dynamically compose them for each input.<n>The Router is optimized via reinforcement learning under task-level rewards, activating latent cognitive primitives in an emergent and compositional manner.
  arXiv  Detail & Related papers  (2026-01-14T08:04:33Z)
• Structured Uncertainty guided Clarification for LLM Agents [126.26213027785813]
  LLM agents extend large language models with tool-calling capabilities, but ambiguous user instructions often lead to incorrect invocations and task failures.<n>We introduce a principled formulation of structured uncertainty over tool-call parameters, modeling joint tool-argument clarification as a POMDP with Expected Value of Perfect Information (EVPI) objective for optimal question selection and aspect-based cost modeling to prevent redundancy.<n>Our SAGE-Agent leverages this structured uncertainty to achieve superior efficiency: increasing coverage on ambiguous tasks by 7-39% while reducing clarification questions by 1.5-2.7$times$ compared to strong prompting and uncertainty-based baselines.
  arXiv  Detail & Related papers  (2025-11-11T21:50:44Z)
• GRIN: GRadient-INformed MoE [132.87651078514122]
  Mixture-of-Experts (MoE) models scale more effectively than dense models due to sparse computation through expert routing. We introduce GRIN (GRadient-INformed MoE training), which incorporates sparse gradient estimation for expert routing. Our model, with only 6.6B activated parameters, outperforms a 7B dense model and matches the performance of a 14B dense model trained on the same data.
  arXiv  Detail & Related papers  (2024-09-18T17:00:20Z)
• Building Math Agents with Multi-Turn Iterative Preference Learning [56.71330214021884]
  This paper studies the complementary direct preference learning approach to further improve model performance.<n>Existing direct preference learning algorithms are originally designed for the single-turn chat task.<n>We introduce a multi-turn direct preference learning framework, tailored for this context.
  arXiv  Detail & Related papers  (2024-09-04T02:41:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.