# Safety-Flag: A Unified Benchmark for the Reliability and Calibration of LLM Content Moderators

arXiv: https://arxiv.org/abs/2609.19072 · HTML: https://mlsanity.com/paper/2609.19072/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 13 of its 27 references that are in the map.
- First posted on arXiv 6 Aug 2026.

## Builds on (most cited first, 13 of 13)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [Language Models (Mostly) Know What They Know](https://mlsanity.com/paper/2207.05221/) (2022)
- [Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs](https://mlsanity.com/paper/2306.13063/) (2023)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs](https://mlsanity.com/paper/2406.18495/) (2024)
- [A Holistic Approach to Undesired Content Detection in the Real World](https://mlsanity.com/paper/2208.03274/) (2022)
- [Mitigating LLM Hallucinations via Conformal Abstention](https://mlsanity.com/paper/2405.01563/) (2024)
- [Just Say No: Analyzing the Stance of Neural Dialogue Generation in Offensive Contexts](https://mlsanity.com/paper/2108.11830/) (2021)
- [Uncertainty-Aware Reliable Text Classification](https://mlsanity.com/paper/2107.07114/) (2021)
- [On Calibration of LLM-based Guard Models for Reliable Content Moderation](https://mlsanity.com/paper/2410.10414/) (2024)
- [LLM Performance Predictors: Learning When to Escalate in Hybrid Human-AI Moderation Systems](https://mlsanity.com/paper/2601.07006/) (2026)

## Built on by (most cited first, 2 of 2)

- [HakemBench: A Turkish Benchmark of Typed Decisions](https://mlsanity.com/paper/2610.02293/) (2026)
- [Calibration and Risk-Budgeted Decisions with System One Models](https://mlsanity.com/paper/2610.19811/) (2026)

## Abstract (arXiv)

Large language models are increasingly used for content moderation, but most evaluations still report aggregate accuracy on individual benchmarks. We introduce Safety-Flag, which places seven widely used safety benchmarks (BeaverTails, XSTest, Ethics, WildGuard, Aegis, ToxiChat, and ToxiGen) into a single balanced flag / do-not-flag protocol. We release item-level decisions and confidence scores for six general-purpose LLMs and four dedicated guards, together with three reference models, evaluated on the same items. Safety-Flag measures three dimensions of moderator reliability: error direction, probability calibration, and confidence-based error ranking for human review. They often disagree. Aggregate accuracy does not reveal error direction: one model flags 85% of benign content, whereas another misses 54% of harmful content. All six general-purpose models are overconfident; fitting one temperature per model reduces calibration error by 2.8--6.0× without changing predicted labels or confidence ordering. Confidence-based abstention lowers selective risk for every model, although the gains depend on how well confidence ranks errors. Dedicated guards produce fewer false alarms and are better calibrated, but several have higher miss rates outside their documented coverage. We release the benchmark, fixed item lists, evaluation code, per-item model outputs, and leaderboard at: https://github.com/yibo-hu-lab/safety-flag-benchmark.