mlsanityData to 7 Oct 2026

When the Model Said 'No Comment', We Knew Helpfulness Was Dead, Honesty Was Alive, and Safety Was Terrified

arXiv 2602.07381 ↗First posted 7 Feb 2026cs.CLcited by 1

On the map: Mixture-of-experts → MoE large language models

Abstract (arXiv)

Large Language Models (LLMs) need to be in accordance with human values-being helpful, harmless, and honest (HHH)-is important for safe deployment. Existing works use Supervised Fine-Tuning (SFT) and Mixture-of-Experts (MoE) to align LLMs. However, these works face challenges in multi-objective settings, such as SFT leading to interference between conflicting objectives, while MoEs suffer from miscalibrated routing. We term this failure mode Axis Collapse, marked by (1) disjoint feature spaces causing catastrophic forgetting, and (2) unreliable inference from misrouted experts. To resolve this, we propose AlignX, a two-stage framework. Stage 1 uses prompt-injected fine-tuning to extract axis-specific task features, mitigating catastrophic forgetting. Stage 2 deploys a MoCaE module that calibrates expert routing using fractal and natural geometry, improving inference reliability. AlignX achieves significant gains on Alpaca (Helpfulness), BeaverTails (Harmlessness), and TruthfulQA (Honesty), with +171.5% win rate, +110.1% in truthfulness-informativeness, and 4.3% fewer safety violations. It also reduces latency and memory usage by over 35% compared to prior MoEs. Results across four LLMs validate its generalizability.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in MoE large language models, one of 12 directions in Mixture-of-experts. Builds on 25 of its 35 references that are in the map. First posted on arXiv 7 Feb 2026.

Builds on

25 of 35 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025Too Helpful, Too Harmless, Too Honest or Just Right?10 Sep
  2. 2024H3Fusion: Helpful, Harmless, Honest Fusion of Aligned LLMs26 Nov
  3. 2020GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding30 Jun

Other references · 22

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2022Training language models to follow instructions with human feedbackcited by 15,712
  3. 2017On Calibration of Modern Neural Networkscited by 4,860
  4. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  5. 2021Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsitycited by 3,014
  6. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,919
  7. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,259
  8. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 828
  9. 2023Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignmentcited by 394
  10. 2024Aligner: Efficient Alignment by Learning to Correctcited by 96
  11. 2024Booster: Tackling Harmful Fine-tuning for Large Language Models via Attenuating Harmful Perturbationcited by 92
  12. 2023Aligning Large Language Models with Human Preferences through Representation Engineeringcited by 71
10 more
  1. 2024(A)I Am Not a Lawyer, But...: Engaging Legal Experts towards Responsible LLM Policies for Legal Advicecited by 61
  2. 2024Analyzing and Reducing Catastrophic Forgetting in Parameter Efficient Tuningcited by 59
  3. 2025Whoever Started the Interference Should End It: Guiding Data-Free Model Merging via Task Vectorscited by 59
  4. 2024Optimizing RLHF Training for Large Language Models with Stage Fusioncited by 41
  5. 2024Knowledge Composition using Task Vectors with Learned Anisotropic Scalingcited by 37
  6. 2024Read-ME: Refactorizing LLMs as Router-Decoupled Mixture of Experts with System Co-Designcited by 18
  7. 2025Should LLM Safety Be More Than Refusing Harmful Instructions?cited by 6
  8. 2024Fractal Calibration for long-tailed object detectioncited by 5
  9. 2025SHIELD: Classifier-Guided Prompting for Robust and Safer LVLMscited by 5
  10. 2025Improving Deep Ensembles by Estimating Confusion Matricescited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026AlignCultura: Towards Culturally Aligned Large Language Models?cited by 1

Similar, not cited

Close in wording, no citation either way

  1. 2025We Think, Therefore We Align LLMs to Helpful, Harmless and Honest Before They Go Wrong
  2. 2023Alignment for Honesty
  3. 2025Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges
  4. 2025DxHF: Providing High-Quality Human Feedback for LLM Alignment via Interactive Decomposition
  5. 2024Improving the Robustness of Large Language Models via Consistency Alignment
  6. 2026Emergent Unfaithfulness: How Alignment Training Causes Language Models to Silently Override Task Faithfulness
  7. 2026SHARD: Safe and Helpful Alignment via Self-Reframing Distillation
  8. 2023Peering Through Preferences: Unraveling Feedback Acquisition for Aligning Large Language Models