The paper "Evading Data Contamination Detection for Language Models is (too) Easy" by Jasper Dekoninck, Mark Niklas Müller, Maximilian Baader, Marc Fischer, and Martin Vechev discusses the issue of data contamination in large language models (LLMs) and the potential for malicious actors to exploit this to improve model performance while evading detection methods. The authors argue that current contamination detection methods are insufficient to address the threat of deliberate contamination by malicious model providers. They propose a categorization of model providers and contamination detection methods, revealing vulnerabilities that can be exploited through *Evasive Augmentation Learning* (EAL). EAL involves rephrasing benchmark samples during fine-tuning to increase model performance while evading detection. The paper demonstrates that EAL can significantly improve benchmark performance by up to 15% while evading all current detection methods. The authors conclude that the integrity of public benchmarks is at risk and call for more robust evaluation methods to address the issue.The paper "Evading Data Contamination Detection for Language Models is (too) Easy" by Jasper Dekoninck, Mark Niklas Müller, Maximilian Baader, Marc Fischer, and Martin Vechev discusses the issue of data contamination in large language models (LLMs) and the potential for malicious actors to exploit this to improve model performance while evading detection methods. The authors argue that current contamination detection methods are insufficient to address the threat of deliberate contamination by malicious model providers. They propose a categorization of model providers and contamination detection methods, revealing vulnerabilities that can be exploited through *Evasive Augmentation Learning* (EAL). EAL involves rephrasing benchmark samples during fine-tuning to increase model performance while evading detection. The paper demonstrates that EAL can significantly improve benchmark performance by up to 15% while evading all current detection methods. The authors conclude that the integrity of public benchmarks is at risk and call for more robust evaluation methods to address the issue.