AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Auditing Black-Box LLM APIs with a Rank-Based Uniformity Test

Xiaoyuan Zhu, Yaowen Ye, Tianyi Alex Qiu, Hanlin Zhu, Sijun Tan, Ajraf Mannan, Jonathan Michala, Raluca Ada Popa, Willie Neiswanger

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 14 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use RUT instead of MMDMMD · Gemma-2-9B-it 4-bit quantization (WildChat)statistical power AUC: 0.214 -> 0.392Improved
Use RUT instead of MMDMMD · Mistral 4-bit quantization (WildChat)statistical power AUC: 0.500 -> 0.586Improved
Use RUT instead of MMDMMD · Llama 8-bit quantization (WildChat)statistical power AUC: 0.158 -> 0.132Hurt
Use RUT instead of KSKS · Llama-3.2-3B 4-bit quantization (WildChat)statistical power AUC: 0.474 -> 0.642Improved
Use RUT instead of MMDMMD · Mistral jailbreak Dan promptstatistical power AUC: 0.802 -> 0.895Improved
Use RUT instead of MMDMMD · Gemma jailbreak Evil-Bot promptstatistical power AUC: 0.753 -> 0.893Improved
Use RUT instead of MMDMMD · Llama 4-bit quantization (Math)statistical power AUC: 0.468 -> 0.532Improved
Use RUT instead of MMDMMD · Gemma 4-bit quantization (BigCodeBench)statistical power AUC: 0.170 -> 0.136Hurt
Use RUT instead of MMDMMD · Llama 4-bit quantization (BigCodeBench)statistical power AUC: 0.353 -> 0.593Improved
Use 8-bit instead of 4-bit quantization as the substituteRUT · Gemma quantization detection (WildChat)statistical power AUC: 0.392 -> 0.049Hurt
Restore missing leading space in HF Inference output (ablation)MMD · Mistral via HF Inferencestatistical power AUC: 1.00 -> 0.211Improved
Query reference with 100 samples per prompt (RUT vs MMD 10 prompts)RUT · Mistral via HF Inferencestatistical power AUC: 0.188No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy