# When No Benchmark Exists: Validating Comparative LLM Safety Scoring Without Ground-Truth Labels

arXiv: https://arxiv.org/abs/2605.06652 · HTML: https://mlsanity.com/paper/2605.06652/

## 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 24 of its 30 references that are in the map.
- First posted on arXiv 7 May 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, 24 of 24)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [A Survey on LLM-as-a-Judge](https://mlsanity.com/paper/2411.15594/) (2024)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Discovering Language Model Behaviors with Model-Written Evaluations](https://mlsanity.com/paper/2212.09251/) (2022)
- [Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned](https://mlsanity.com/paper/2209.07858/) (2022)
- [Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge](https://mlsanity.com/paper/2406.07791/) (2024)
- [SafetyBench: Evaluating the Safety of Large Language Models](https://mlsanity.com/paper/2309.07045/) (2023)
- [AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions](https://mlsanity.com/paper/2506.09038/) (2025)
- [Large Language Models Often Know When They Are Being Evaluated](https://mlsanity.com/paper/2505.23836/) (2025)
- [Measuring what Matters: Construct Validity in Large Language Model Benchmarks](https://mlsanity.com/paper/2511.04703/) (2025)
- [Measurement to Meaning: A Validity-Centered Framework for AI Evaluation](https://mlsanity.com/paper/2505.10573/) (2025)
- [AILuminate: Introducing v1.0 of the AI Risk and Reliability Benchmark from MLCommons](https://mlsanity.com/paper/2503.05731/) (2025)
- [Probing and Steering Evaluation Awareness of Language Models](https://mlsanity.com/paper/2507.01786/) (2025)
- [NorBench -- A Benchmark for Norwegian Language Models](https://mlsanity.com/paper/2305.03880/) (2023)
- [LinguaSafe: A Comprehensive Multilingual Safety Benchmark for Large Language Models](https://mlsanity.com/paper/2508.12733/) (2025)
- [Comparison requires valid measurement: Rethinking attack success rate comparisons in AI red teaming](https://mlsanity.com/paper/2601.18076/) (2026)
- [NLEBench+NorGLM: A Comprehensive Empirical Analysis and Benchmark Dataset for Generative Language Models in Norwegian](https://mlsanity.com/paper/2312.01314/) (2023)
- [NorEval: A Norwegian Language Understanding and Generation Evaluation Benchmark](https://mlsanity.com/paper/2504.07749/) (2025)
- [UK AISI Alignment Evaluation Case-Study](https://mlsanity.com/paper/2604.00788/) (2026)
- [CyclicJudge: Mitigating Judge Bias Efficiently in LLM-based Evaluation](https://mlsanity.com/paper/2603.01865/) (2026)

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

- [False Floors: LLM Safety Routing Evaluations Break Under Distribution Shift](https://mlsanity.com/paper/2610.01535/) (2026)

## Abstract (arXiv)

Many deployments must compare candidate language models for safety before a labeled benchmark exists for the relevant language, sector, or regulatory regime. We formalize this setting as benchmarkless comparative safety scoring and specify the contract under which a scenario-based audit can be interpreted as deployment evidence. Scores are valid only under a fixed scenario pack, rubric, auditor, judge, sampling configuration, and rerun budget. Because no labels are available, we replace ground-truth agreement with an instrumental-validity chain: responsiveness to a controlled safe-versus-abliterated contrast, dominance of target-driven variance over auditor and judge artifacts, and stability across reruns. We instantiate the chain in SimpleAudit, a local-first scoring instrument, and validate it on a Norwegian safety pack. Safe and abliterated targets separate with AUROC values between 0.89 and 1.00, target identity is the dominant variance component (η2 ≈ 0.52), and severity profiles stabilize by ten reruns. Applying the same chain to Petri shows that it admits both tools. The substantial differences arise upstream of the chain, in claim-contract enforcement and deployment fit. A Norwegian public-sector procurement case comparing Borealis and Gemma 3 demonstrates the resulting evidence in practice: the safer model depends on scenario category and risk measure. Consequently, scores, matched deltas, critical rates, uncertainty, and the auditor and judge used must be reported together rather than collapsed into a single ranking.