mlsanityData to 7 Oct 2026

Making Your LLMs More Objective: Stabilizing LLM Safety Behavior Across Traits with Trait-Invariant Safety Tuning

arXiv 2608.11705 ↗First posted 12 Aug 2026cs.AIcited by 0

On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks

also in 1 more
Abstract (arXiv)

Aligned large language models (LLMs) are expected to exhibit safety behavior based on the content of the user request: they should refuse unsafe requests and comply with safe ones. However, we show that the same request can elicit substantially different safety decisions under different traits assigned in the system prompt, a failure mode we call trait-induced safety variation. To measure this failure, we introduce refusal-based metrics: Trait-Induced Deviation measures dataset-level deviation from the no-trait baseline, while Trait-Induced Flip Rate measures whether the same request receives different safety decisions across traits. We then provide a representation-level analysis of the mechanism behind trait-induced safety shifts and find that traits perturb the model's safety representations within a low-dimensional subspace. To achieve trait-invariant safety, where safety behavior remains stable across traits, we introduce Trait-Invariant Safety Tuning (TIST), a simple yet effective self-distillation framework that aligns an LLM's trait-conditioned behavior with its no-trait behavior. Guided by our analysis, we further propose Trait-Subspace Neutralization (TraSN), an instantiation of TIST, which enforces invariance only within the identified trait subspace. Experiments show that TraSN improves trait-invariant safety and strengthens harmful-request safety while preserving general capability. Our results highlight traits as an important factor in LLM safety and robust model behavior.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 38 of its 44 references that are in the map. First posted on arXiv 12 Aug 2026.

Builds on

38 of 44 references.

Closest ancestors · 7

Same direction, newest first

  1. 2026Deactivating Refusal Triggers: Understanding and Mitigating Overrefusal in Safety Alignment12 Mar
  2. 2026Fail-Closed Alignment for Large Language Models19 Feb
  3. 2026Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control7 Feb
  4. 2025Mitigating Over-Refusal in Aligned Large Language Models via Inference-Time Activation Energy9 Oct
  5. 2025Beyond Surface Alignment: Rebuilding LLMs Safety Mechanism via Probabilistically Ablating Refusal Direction18 Sep
  6. 2024PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference20 Jun
  7. 2024Safety Alignment Should Be Made More Than Just a Few Tokens Deep10 Jun

Other references · 31

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2024The Llama 3 Herd of Modelscited by 14,192
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  4. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  5. 2023Let's Verify Step by Stepcited by 3,598
  6. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  7. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,143
  8. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 960
  9. 2024JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Modelscited by 494
  10. 2024WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMscited by 460
  11. 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 441
  12. 2023Catastrophic Jailbreak of Open-source LLMs via Exploiting Generationcited by 439
19 more
  1. 2024Scaling Synthetic Data Creation with 1,000,000,000 Personascited by 397
  2. 2024A StrongREJECT for Empty Jailbreakscited by 352
  3. 2025Persona Vectors: Monitoring and Controlling Character Traits in Language Modelscited by 303
  4. 2024Improving Alignment and Robustness with Circuit Breakerscited by 302
  5. 2024WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Modelscited by 206
  6. 2026Gemma 4 Technical Reportcited by 187
  7. 2026Qwen3.5-Omni Technical Reportcited by 146
  8. 2023ExpertPrompting: Instructing Large Language Models to be Distinguished Expertscited by 140
  9. 2025Persona Features Control Emergent Misalignmentcited by 76
  10. 2024Do LLMs Have Distinct and Consistent Personality? TRAIT: Personality Testset designed for LLMs with Psychometricscited by 56
  11. 2024Surgical, Cheap, and Flexible: Mitigating False Refusal in Language Models via Single Vector Ablationcited by 48
  12. 2023Learn to Refuse: Making Large Language Models More Controllable and Reliable through Knowledge Scope Limitation and Refusal Mechanismcited by 33
  13. 2024Multi-expert Prompting Improves Reliability, Safety, and Usefulness of Large Language Modelscited by 13
  14. 2025Think Before Refusal : Triggering Safety Reflection in LLMs to Mitigate False Refusal Behaviorcited by 12
  15. 2026Expert Personas Improve LLM Alignment but Damage Accuracy: Bootstrapping Intent-Based Persona Routing with PRISMcited by 9
  16. 2025No for Some, Yes for Others: Persona Prompts and Other Sources of False Refusal in Language Modelscited by 8
  17. 2026When Does Persona Prompting Actually Help? A Retrieval and Metric Analysis of Expert Role Injection in LLMscited by 4
  18. 2026Tracing Persona Vectors Through LLM Pretrainingcited by 3
  19. 2026Persona Jailbreaking in Large Language Modelscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning
  2. 2026Understanding and Preserving Safety in Fine-Tuned LLMs
  3. 2025The Hidden Dimensions of LLM Alignment: A Multi-Dimensional Analysis of Orthogonal Safety Directions
  4. 2024Locking Down the Finetuned LLMs Safety
  5. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!
  6. 2025Picky LLMs and Unreliable RMs: An Empirical Study on Safety Alignment after Instruction Tuning
  7. 2025Fine-Tuning Lowers Safety and Disrupts Evaluation Consistency
  8. 2026From Parameter Dynamics to Risk Scoring : Quantifying Sample-Level Safety Degradation in LLM Fine-tuning