Data Contamination in Language Models

Follow this topic Ask about this topic

Momentum

4 papers in the last four weeks, level with the four weeks before. 0.0% of all new papers.

Jul 13Week of Sep 28

Latest papers 39

All topics

Correct, Don't Delete: Mitigating Emergent Misalignment with Corrective Supervision
Sep 29, 2026Jacob EpifanoLLM Alignment Emergent Misalignment in Language Models
Independent researcher
Uncheatable Eval: Dynamic Compression-Based Evaluation of Language Models
Sep 23, 2026Kaifeng Tan, Yudong Li, Linlin ShenLLM Evaluation Data Contamination in Language Models
Shenzhen University · Tsinghua University
Mirror, Mirror on the Wall: Prompt Echoing in Small Instruct Language Models
Sep 14, 2026Inez Okulska, Bartosz Naskręcki, Jan Piotrowski +1Data Contamination in Language Models Induction Heads
Centre for Credible AI, Warsaw University of Technology · Adam Mickiewicz University Poznan · Institute of Fundamental Technological Research, Polish Academy of Sciences
SynthSentry: Detecting Synthetic Data Contamination in Language Model Training Data
Sep 14, 2026Praveen Kumar Myakala, Ravichandra Namburi, Sowmya Keragodu Jayaramu +1AI-Generated Text Detection Training Data Curation
Independent AI Researcher Melissa, USA · Independent Researcher Frisco, USA · Independent Researcher Austin, USA
Molecular Déjà Vu: Digit-Level Retrieval of Molecular Properties in Frontier Language Models
Sep 4, 2026Matthias Busch, Marius Tacke, Sviatlana V. Lamaka +4LLM Evaluation LLM Reliability
Institute for Artificial Intelligence and Simulation in Mechanics, Hamburg University of Technology, Eißendorfer Straße, 21073 Hamburg, Germany · Institute of Material Systems Modeling, Helmholtz-Zentrum Hereon, Max-Planck-Straße, 21502 Geesthacht, Germany · Institute of Surface Science, Helmholtz-Zentrum Hereon, Max-Planck-Straße, 21502 Geesthacht, Germany +3
Excess Separability: Nuisance-Controlled Residual-Stream Probing for Benchmark Contamination Detection
Aug 12, 2026Florian BraunBenchmark Contamination Data Contamination in Language Models
Calibrating Post-Training Feature Shifts for LLM Data Contamination Detection
Aug 11, 2026Zhen Yang, Mengqi Wang, Gengda Zhao +3Membership Inference Attacks Data Contamination in Language Models
The University of New South Wales
Cultivar: A Contrastive and Locale-Oriented Translation Benchmark for Investigating Contamination and Localisation Robustness
Aug 10, 2026Pinzhen Chen, Koel Dutta Chowdhury, Xiaoya Xu +20Data Contamination in Language Models Language Model Robustness
Queen’s University Belfast · University of Technology Nuremberg · Saarland University +12
Zero Gap Is Not Restoration: Stratified Per-Question Probability Evaluation and Step-wise Mitigation of Benchmark Contamination
Aug 7, 2026Ruijie Hou, Yueyang Jiao, Zhao Wang +1LLM Evaluation Memorization in Generative Models
Zhejiang University
Library Reachability in LSR-Synth: How Anti-Memorization Design Changes the Measurement of Symbolic Discovery
Jul 30, 2026Zhan'ao Yao, Liang Yin, Zhihao Gao +9Synthetic Benchmark Generation Data Contamination in Language Models
State Key Laboratory of High Performance Ceramics, Shanghai Institute of Ceramics, Chinese Academy of Sciences, 1295 Dingxi Road, Shanghai 200050, China · Center of Materials Science and Optoelectronics Engineering, University of Chinese Academy of Sciences, Beijing 100049, China · School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences, Beijing, China +4
How Context Attribution Handles What the Model Already Knows
Jul 26, 2026Quoc-Huy Trinh, Lin Zhu, Sebastian SzyllerLLM Evaluation LLM Interpretability
Aalto University
Novel Claim or Déjà Vu? Rethinking "Contamination-Free'' Dynamic Evaluation for Multimodal Automated Fact-Checking
Jul 26, 2026Haorui He, Xinwen Chen, Dacheng Wen +3Benchmark Contamination Automated Fact-Checking
Department of Interactive Media, Hong Kong Baptist University · School of Computing and Data Science, The University of Hong Kong · Faculty of Science and Technology, Beijing Normal-Hong Kong Baptist University
Spaghetti Architect: A Contamination-Resistant, By-Construction-Labelled, Multi-Language Code Dataset Generator
Jul 21, 2026Yuxiang JiSynthetic Benchmark Generation Data Contamination in Language Models
Sunway College
RustMizan: A Compilable, Contamination-Aware Benchmarking Framework for Rust Vulnerabilities
Jul 6, 2026Tarek Elsayed, Shiping Yang, Eunsong Koh +13Software Vulnerability Detection Data Contamination in Language Models
Simon Fraser University · Trinity University · Microsoft Research +1
Defending Against Harmful Supervision Hidden in Benign Samples
Jun 29, 2026Bang An, Yibo Yang, Dandan Guo +3Supervised Fine-Tuning Fine-Tuning
1King Abdullah University of Science and Technology · School of Artificial Intelligence, Jilin University
SrDetection: A Self-Referential Framework for Data Leakage Detection in Code Large Language Models
Jun 29, 2026Shuaimin Li, Liyang Fan, Zeyang Li +9LLM Evaluation LLM Auditing
Shenzhen Key Laboratory for High Performance Data Mining, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences · Shenzhen University · Artificial Intelligence Research Institute, Shenzhen University of Advanced Technology +5
Uncertainty-based Debiasing and Unlearning for Decontamination
Jun 22, 2026Guangzhi Sun, Xiao Zhan, Mark GalesData Contamination in Language Models LLM Unlearning
Department of Engineering, University of Cambridge · VRAIN, Universitat Politècnica de València
Calibration Without Comprehension: Diagnosing the Limits of Fine-Tuning LLMs for Vulnerability Detection in Systems Software
Jun 18, 2026Arastoo Zibaeirad, Marco VieiraLLM Evaluation Software Vulnerability Detection
University of North Carolina at Charlotte, Charlotte, NC, USA
Dango: A Strictly L1-Only Large Language Model for Studying Second Language Acquisition
Jun 17, 2026Shiho Matta, Yin Jou Huang, Fei Cheng +3Language Model Pretraining Data Contamination in Language Models
Kyoto University · NII-LLMC
CANARY: Zero-Label Detection of Fine-Tuning Contamination in Language Models
Jun 1, 2026Swapnil ParekhLLM Auditing Data Contamination in Language Models
Intuit Inc.
Subtitle-Aligned Fine-Tuning of Whisper for Swiss German ASR: Benchmark Contamination, Convention Mismatch, and an Honest Baseline at 25.6% WER (13.8% cWER)
May 29, 2026Felix AkeretASR Evaluation Fine-Tuning
Independent Researcher, Zurich, Switzerland · Dipl. Ing. ETH Zürich, Dipl. NDS BWI ETH Zürich. · Dipl. NDS BWI ETH Zürich.
LaRA: Layer-wise Representation Analysis for Detecting Data Contamination in RL Post-Training
May 28, 2026Minju Gwak, Minseo Kwak, Dongseok Lee +3Data Contamination in Language Models RL Post-Training
Yonsei University · Seoul National University · Georgia Institute of Technology
NumLeak: Public Numeric Benchmarks as Latent Labels in Foundation Models
May 28, 2026Anany KotawalaData Leakage LLM Evaluation
Princeton University, Princeton, NJ, USA.
MemGuard: Preventing Memory Contamination in Long-Term Memory-Augmented Large Language Models
May 27, 2026Hyeonjeong Ha, Jeonghwan Kim, Cheng Qian +7Memory-Augmented Language Models LLM Hallucination Mitigation
University of Illinois Urbana-Champaign · Capital One · Columbia University
MAIGO: Mitigating Lost-in-Conversation with History-Cleaned On-Policy Self-Distillation
May 26, 2026Haoyu Zheng, Yun Zhu, Shu Yuan +5Instruction Following Data Contamination in Language Models
1Zhejiang University · 2Shanghai AI Laboratory · 3Huazhong University of Science and Technology +2
Cordyceps: Covert Control Attacks on LLMs via Data Poisoning
May 26, 2026Zedian Shao, Charles Fleming, Teodora BalutaLLM Backdoor Attacks Adversarial Attacks on LLMs
Georgia Institute of Technology · Cisco Systems
TRACER: A Semantic-Aware Framework for Fine-Grained Contamination Detection in Code LLMs
May 22, 2026Yifeng Di, Xuliang Huang, Tianyi ZhangLLM Auditing Data Contamination in Language Models
Purdue University West Lafayette, IN · University of Illinois Urbana-Champaign Champaign, IL
Pretraining Data Exposure in Large Language Models: A Survey of Membership Inference, Data Contamination, and Security Implications
May 21, 2026Ziyi Tong, Feifei Sun, Le Minh NguyenPrivacy Leakage in Language Models Data Contamination in Language Models
Japan Advanced Institute of Science and Technology, Japan
The Illusion of Reasoning: Exposing Evasive Data Contamination in LLMs via Zero-CoT Truncation
May 21, 2026Yifan Lan, Yuanpu Cao, Hanyu Wang +2LLM Evaluation LLM Auditing
The Pennsylvania State University
Provable Joint Decontamination for Benchmarking Multiple Large Language Models
May 20, 2026Zhenlong Liu, Hao Zeng, Hongxin WeiLLM Evaluation Benchmark Contamination
Department of Statistics and Data Science, Southern University of Science and Technology · Shanghai Innovation Institute

Sep 29, 2026, cs.CLJ/K move · Enter open · S save

Correct, Don't Delete: Mitigating Emergent Misalignment with Corrective Supervision

Jacob Epifano

Independent researcher

LLM Alignment Emergent Misalignment in Language Models Data Contamination in Language Models

Open paper AskPDF arXiv

Fine-tuning a language model on a narrow set of harmful demonstrations, such as bad medical advice, can make it broadly misaligned on unrelated questions, a phenomenon known as emergent misalignment (EM). The usual defense is to find the offending rows and delete them, but a row locator failed our held-out test and deleting rows helps less than expected. We ask a different question: given a fixed set of poisoned rows, is it better to correct them than to remove them? We fine-tune Qwen2.5-14B-Instruct on a mixture of bad medical advice and benign chat data, select a quarter of the poison rows in advance, and either delete them or replace each with a corrected answer to the same prompt, keeping everything else the same. Replacing the rows cuts the EM rate by about a third and improves answers on held-out medical questions, while deleting the same rows has little measurable effect. The advantage is larger when half the poison rows are corrected, and it holds on a second base model and a second misaligned model organism. The content of the replacement appears to matter: paraphrasing the rows while keeping their bad advice shows no clear benefit, and the correct answers distributed with the dataset appear to do about as well as our rewriter's. Realigning an already-poisoned model with further fine-tuning is known to work, but which data does the work has not been compared directly. We find that a short round of training on corrections beats the same amount of training on generic chat data, that corrections on other medical prompts do roughly as well as corrections of the poisoned prompts themselves, and that instructing the correction writer to model a careful, harm-avoiding assistant adds no measurable benefit over plain corrections. In the settings we tested, correcting harmful training data reduces EM more than deleting it.

Momentum

Related topics

Latest papers 39