mlsanityData to 7 Oct 2026

Mixture-of-experts routing safety

20 papers1 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: routing, moe, experts, expert, safety

Key facts. Mixture-of-experts routing safety: 20 papers on arXiv since 2025 (data through 7 Oct 2026). 9 of its 20 mapped papers (45%) cite “BadMoE: Backdooring Mixture-of-Experts LLMs via Optimizing Routing Triggers and Infecting Dormant Experts” (2025, arXiv 2504.18598). 9 of its 20 mapped papers (45%) cite “SAFEx: Analyzing Vulnerabilities of MoE-Based LLMs via Stable Safety-critical Expert Identification” (2025, arXiv 2506.17368). 1 new paper 24 Sep–7 Oct.

What’s new

24 Sep – 7 Oct 2026

1 new since 24 Sep

  1. When Routing Reveals Membership: Privacy Leakage from MoE Router Telemetry

Foundations

Most cited by papers here

  1. BadMoE: Backdooring Mixture-of-Experts LLMs via Optimizing Routing Triggers and Infecting Dormant Experts2025cited by 9 here
  2. SAFEx: Analyzing Vulnerabilities of MoE-Based LLMs via Stable Safety-critical Expert Identification2025cited by 9 here
  3. Steering MoE LLMs via Expert (De)Activation2025cited by 8 here
  4. GateBreaker: Gate-Guided Attacks on Mixture-of-Expert LLMs2025cited by 4 here
  5. Sparse Models, Sparse Safety: Unsafe Routes in Mixture-of-Experts LLMs2026cited by 3 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 4 here
  2. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 13 here
  3. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 5 here
  4. 2023Catastrophic Jailbreak of Open-source LLMs via Exploiting Generationcited by 2 here
  5. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 6 here
  6. 2024How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMscited by 5 here
  7. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 4 here
  8. 2024A StrongREJECT for Empty Jailbreakscited by 8 here
  9. 2024JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Modelscited by 3 here
  10. 2025TwinBreak: Jailbreaking LLM Security Alignments based on Twin Promptscited by 3 here
  11. 2025GateBreaker: Gate-Guided Attacks on Mixture-of-Expert LLMscited by 4 here
13 more it builds on, not in this chain
  1. 2023"Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Modelscited by 3 here
  2. 2023Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignmentcited by 3 here
  3. 2023GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Promptscited by 2 here
  4. 2023AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Modelscited by 5 here
  5. 2023DeepInception: Hypnotize Large Language Model to Be Jailbreakercited by 3 here
  6. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Promptscited by 2 here
  7. 2024Safety Alignment Should Be Made More Than Just a Few Tokens Deepcited by 5 here
  8. 2024SEUF: Is Unlearning One Expert Enough for Mixture-of-Experts LLMs?cited by 3 here
  9. 2025BadMoE: Backdooring Mixture-of-Experts LLMs via Optimizing Routing Triggers and Infecting Dormant Expertscited by 9 here
  10. 2025SAFEx: Analyzing Vulnerabilities of MoE-Based LLMs via Stable Safety-critical Expert Identificationcited by 9 here
  11. 2025Steering MoE LLMs via Expert (De)Activationcited by 8 here
  12. 2025NeuroStrike: Neuron-Level Attacks on Aligned LLMscited by 4 here
  13. 2026Sparse Models, Sparse Safety: Unsafe Routes in Mixture-of-Experts LLMscited by 3 here

The 24 most cited of 116.

Browse

15 of 20, ranked 6–20

  1. 6Large Language Lobotomy: Jailbreaking Mixture-of-Experts via Expert Silencing20262
  2. 7Life-Cycle Routing Vulnerabilities of LLM Router20252
  3. 8Routing-Aware Safety Alignment for Mixture-of-Experts Models20261
  4. 9Route to Rome Attack: Directing LLM Routers to Expensive Models via Adversarial Suffix Optimization20261
  5. 10How Robust Are Router-LLMs? Analysis of the Fragility of LLM Routing Capabilities20251
  6. 11GRIP: Algorithm-Agnostic Machine Unlearning for Mixture-of-Experts via Geometric Router Constraints20260
  7. 12RerouteGuard: Understanding and Mitigating Adversarial Risks for LLM Routing20260
  8. 13The Workload-Router-Pool Architecture for LLM Inference Optimization: A Vision Paper from the vLLM Semantic Router Project20260
  9. 14RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs20260
  10. 15Misrouter: Exploiting Routing Mechanisms for Input-Only Attacks on Mixture-of-Experts LLMs20260
All 15
  1. 16Safety-Oriented Routing Analysis of Mixtral MoE Under Benign and Harmful Prompts20260
  2. 17RASET: Router-Agnostic Safety-Critical Expert Tuning Exposes Localized Safety Enforcement Failures in Mixture-of-Experts LLMs20260
  3. 18MESA: Improving MoE Safety Alignment via Decentralized Expertise20260
  4. 19SEAL: Reinforcing Global Safety in Mixture-of-Experts through Shared Expert ALignment20260
  5. 20Who Speaks for the Trigger? Dynamic Expert Routing in Backdoored Mixture-of-Experts Transformers20250