# LLM-Safety Evaluations Lack Robustness

arXiv: https://arxiv.org/abs/2503.02574 · HTML: https://mlsanity.com/paper/2503.02574/

## Key facts

- Cited by 18 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 83 of its 100 references that are in the map.
- First posted on arXiv 4 Mar 2025.

## Placed in

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

## Builds on (most cited first, 30 of 83)

- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [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)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [Explaining and Harnessing Adversarial Examples](https://mlsanity.com/paper/1412.6572/) (2014)
- [Towards Deep Learning Models Resistant to Adversarial Attacks](https://mlsanity.com/paper/1706.06083/) (2017)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context](https://mlsanity.com/paper/2403.05530/) (2024)
- [ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes](https://mlsanity.com/paper/1702.04405/) (2017)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [The Curious Case of Neural Text Degeneration](https://mlsanity.com/paper/1904.09751/) (2019)
- [SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing](https://mlsanity.com/paper/1808.06226/) (2018)
- [Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks](https://mlsanity.com/paper/2003.01690/) (2020)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [Jailbreaking Black Box Large Language Models in Twenty Queries](https://mlsanity.com/paper/2310.08419/) (2023)
- [Representation Engineering: A Top-Down Approach to AI Transparency](https://mlsanity.com/paper/2310.01405/) (2023)
- [Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples](https://mlsanity.com/paper/1802.00420/) (2018)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models](https://mlsanity.com/paper/2310.04451/) (2023)
- [Phi-4 Technical Report](https://mlsanity.com/paper/2412.08905/) (2024)
- [RobustBench: a standardized adversarial robustness benchmark](https://mlsanity.com/paper/2010.09670/) (2020)
- [Safe RLHF: Safe Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2310.12773/) (2023)
- [WildChat: 1M ChatGPT Interaction Logs in the Wild](https://mlsanity.com/paper/2405.01470/) (2024)
- [GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Prompts](https://mlsanity.com/paper/2309.10253/) (2023)

## Built on by (most cited first, 18 of 18)

- [The Scales of Justitia: A Comprehensive Survey on Safety Evaluation of LLMs](https://mlsanity.com/paper/2506.11094/) (2025)
- [A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness](https://mlsanity.com/paper/2603.06594/) (2026)
- [AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents](https://mlsanity.com/paper/2604.02947/) (2026)
- [Robustness in Large Language Models: A Survey of Mitigation Strategies and Evaluation Metrics](https://mlsanity.com/paper/2505.18658/) (2025)
- [Toward Safer Diffusion Language Models: Discovery and Mitigation of Priming Vulnerability](https://mlsanity.com/paper/2510.00565/) (2025)
- [Diffusion LLMs are Natural Adversaries for any LLM](https://mlsanity.com/paper/2511.00203/) (2025)
- [AdversariaLLM: A Unified and Modular Toolbox for LLM Robustness Research](https://mlsanity.com/paper/2511.04316/) (2025)
- [How Sensitive Are Safety Benchmarks to Judge Configuration Choices?](https://mlsanity.com/paper/2604.24074/) (2026)
- [Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem](https://mlsanity.com/paper/2506.17299/) (2025)
- [Beyond Jailbreaks: Revealing Stealthier and Broader LLM Security Risks Stemming from Alignment Failures](https://mlsanity.com/paper/2506.07402/) (2025)
- [PLAGUE: Plug-and-play framework for Lifelong Adaptive Generation of Multi-turn Exploits](https://mlsanity.com/paper/2510.17947/) (2025)
- [Intent Laundering: AI Safety Datasets Are Not What They Seem](https://mlsanity.com/paper/2602.16729/) (2026)
- [Trusted Weights, Treacherous Optimizations? Optimization-Triggered Backdoor Attacks on LLMs](https://mlsanity.com/paper/2605.20641/) (2026)
- [Closing the Distribution Gap in Adversarial Training for LLMs](https://mlsanity.com/paper/2602.15238/) (2026)
- [When Efficiency Backfires: Cascading LLMs Trigger Cascade Failure under Adversarial Attack](https://mlsanity.com/paper/2605.17288/) (2026)
- [Test-Time Training Undermines Safety Guardrails](https://mlsanity.com/paper/2605.22984/) (2026)
- [EvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety Failures](https://mlsanity.com/paper/2606.30219/) (2026)
- [Jailbreaking Open-Weight LLMs via Random Embedding Perturbations](https://mlsanity.com/paper/2610.07125/) (2026)

## Abstract (arXiv)

In this paper, we argue that current safety alignment research efforts for large language models are hindered by many intertwined sources of noise, such as small datasets, methodological inconsistencies, and unreliable evaluation setups. This can, at times, make it impossible to evaluate and compare attacks and defenses fairly, thereby slowing progress. We systematically analyze the LLM safety evaluation pipeline, covering dataset curation, optimization strategies for automated red-teaming, response generation, and response evaluation using LLM judges. At each stage, we identify key issues and highlight their practical impact. We also propose a set of guidelines for reducing noise and bias in evaluations of future attack and defense papers. Lastly, we offer an opposing perspective, highlighting practical reasons for existing limitations. We believe that addressing the outlined problems in future research will improve the field's ability to generate easily comparable results and make measurable progress.