mlsanityData to 14 Oct 2026

MASteer: Multi-Agent Adaptive Steer Strategy for End-to-End LLM Trustworthiness Repair

arXiv 2508.06963 ↗First posted 9 Aug 2025cs.AI cs.LGcited by 0

On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks

also in 1 more
Abstract (arXiv)

Large Language Models (LLMs) face persistent and evolving trustworthiness issues, motivating developers to seek automated and flexible repair methods that enable convenient deployment across diverse scenarios. Existing repair methods like supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) are costly and slow, while prompt engineering lacks robustness and scalability. Representation engineering, which steers model behavior by injecting targeted concept vectors during inference, offers a lightweight, training-free alternative. However, current approaches depend on manually crafted samples and fixed steering strategies, limiting automation and adaptability. To overcome these challenges, we propose MASteer, the first end-to-end framework for trustworthiness repair in LLMs based on representation engineering. MASteer integrates two core components: AutoTester, a multi-agent system that generates diverse, high-quality steer samples tailored to developer needs; and AutoRepairer, which constructs adaptive steering strategies with anchor vectors for automated, context-aware strategy selection during inference. Experiments on standard and customized trustworthiness tasks show MASteer consistently outperforms baselines, improving metrics by 15.36% on LLaMA-3.1-8B-Chat and 4.21% on Qwen-3-8B-Chat, while maintaining general model capabilities. MASteer demonstrates strong robustness, generalization, and practical value for scalable, efficient trustworthiness repair.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 25 of its 31 references that are in the map. First posted on arXiv 9 Aug 2025.

Builds on

25 of 31 references.

References · 25

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2022Training language models to follow instructions with human feedbackcited by 15,971
  3. 2024The Llama 3 Herd of Modelscited by 14,715
  4. 2025Qwen3 Technical Reportcited by 8,820
  5. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,720
  6. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 3,050
  7. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  8. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  9. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,207
  10. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  11. 2024Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluatorscited by 897
  12. 2021BBQ: A Hand-Built Bias Benchmark for Question Answeringcited by 701
13 more
  1. 2023Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignmentcited by 401
  2. 2023Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructionscited by 387
  3. 2023In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steeringcited by 245
  4. 2024ShieldGemma: Generative AI Content Moderation Based on Gemmacited by 243
  5. 2024Programming Refusal with Conditional Activation Steeringcited by 206
  6. 2024Adaptive Activation Steering: A Tuning-Free LLM Truthfulness Improvement Method for Diverse Hallucinations Categoriescited by 77
  7. 2024Spectral Editing of Activations for Large Language Model Alignmentcited by 56
  8. 2024SCANS: Mitigating the Exaggerated Safety for LLMs via Safety-Conscious Activation Steeringcited by 50
  9. 2025A Unified Understanding and Evaluation of Steering Methodscited by 46
  10. 2024BiasAlert: A Plug-and-play Tool for Social Bias Detection in LLMscited by 21
  11. 2024Discovering Bias in Latent Space: An Unsupervised Debiasing Approachcited by 15
  12. 2025SafeSteer: Interpretable Safety Steering with Refusal-Evasion in LLMscited by 10
  13. 2025Guiding Giants: Lightweight Controllers for Weighted Activation Steering in LLMscited by 8

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025A Survey on Trustworthy LLM Agents: Threats and Countermeasures
  2. 2026SkillAdaptor: Self-Adapting Skills for LLM Agents from Trajectories
  3. 2026Better Harnesses, Smaller Models: Building 90% Cheaper Agents via Automated Harness Adaptation
  4. 2026TRUSTMEM: Learning Trustworthy Memory Consolidation for LLM Agents with Long-Term Memory
  5. 2024Tiny Refinements Elicit Resilience: Toward Efficient Prefix-Model Against LLM Red-Teaming
  6. 2026FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents
  7. 2026TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination
  8. 2026SAT: Sequential Agent Tuning for Coordinator Free Plug and Play Multi-LLM Training with Monotonic Improvement Guarantees