# When Refusal Looks Safe: The Refusal-Cue Shortcut in Safety Guard Models

arXiv: https://arxiv.org/abs/2608.03201 · HTML: https://mlsanity.com/paper/2608.03201/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security.
- Builds on 26 of its 30 references that are in the map.
- First posted on arXiv 4 Aug 2026.

## Placed in

- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)

## Builds on (most cited first, 26 of 26)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [Jailbroken: How Does LLM Safety Training Fail?](https://mlsanity.com/paper/2307.02483/) (2023)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization](https://mlsanity.com/paper/1911.08731/) (2019)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [Towards Automated Circuit Discovery for Mechanistic Interpretability](https://mlsanity.com/paper/2304.14997/) (2023)
- [XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models](https://mlsanity.com/paper/2308.01263/) (2023)
- [WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs](https://mlsanity.com/paper/2406.18495/) (2024)
- [Safety Alignment Should Be Made More Than Just a Few Tokens Deep](https://mlsanity.com/paper/2406.05946/) (2024)
- [Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrails](https://mlsanity.com/paper/2501.09004/) (2025)
- [GuardReasoner: Towards Reasoning-based LLM Safeguards](https://mlsanity.com/paper/2501.18492/) (2025)
- [LLMs Encode Harmfulness and Refusal Separately](https://mlsanity.com/paper/2507.11878/) (2025)
- [STAIR: Improving Safety Alignment with Introspective Reasoning](https://mlsanity.com/paper/2502.02384/) (2025)
- [Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training](https://mlsanity.com/paper/2407.09121/) (2024)
- [S-Eval: Towards Automated and Comprehensive Safety Evaluation for Large Language Models](https://mlsanity.com/paper/2405.14191/) (2024)
- [Safety Through Reasoning: An Empirical Study of Reasoning Guardrail Models](https://mlsanity.com/paper/2505.20087/) (2025)
- [AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning](https://mlsanity.com/paper/2507.14987/) (2025)
- [SafeSeek: Universal Attribution of Safety Circuits in Language Models](https://mlsanity.com/paper/2603.23268/) (2026)

## Built on by (most cited first, 1 of 1)

- [Unread or Unenforced? Separating Representation from Enforcement Failure in Content Guards](https://mlsanity.com/paper/2609.26178/) (2026)

## Abstract (arXiv)

Safety guards are widely used to filter harmful content and are typically trained via supervised fine-tuning on labeled prompt-response pairs. We audit two widely used safety-guard training datasets, WildGuardMix and GR-Train, and find that among responses to harmful prompts, refusal expressions co-occur almost exclusively with unharmful labels. This imbalance motivates what we term the refusal-cue shortcut: inserting a refusal cue into a harmful response could flip the guard's verdict from harmful to unharmful. The shortcut affects not only guards trained on these datasets but also officially released models such as LlamaGuard3 and Qwen3Guard whose training data is undisclosed. It persists across response positions and is generally stronger in smaller variants within a family. To mitigate it, we adapt sparse complementary masking as a lightweight post-hoc intervention that identifies and suppresses a small set of shortcut-associated attention heads and MLP neurons without retraining. On two primary benchmarks, the intervention achieves an approximately 79% relative reduction in response-initial detection failures induced by refusal cues, while preserving standard detection performance. Although optimized using cues at a single response position, the suppression effect transfers to unseen positions and datasets, suggesting that shortcut manifestations across positions are partly mediated by shared internal components. Further analysis provides evidence that shortcut reliance and legitimate refusal recognition are partially functionally separable, as suppressing the shortcut broadly preserves the guard's ability to recognize genuine refusals.