DyePack: Provably Flagging Test Set Contamination in LLMs Using Backdoors
Abstract
DyePack, a framework using backdoor attacks, identifies models that leveraged benchmark test sets during training by introducing benign backdoor samples, ensuring precise false positive rates while preventing false accusations.
Open benchmarks are essential for evaluating and advancing large language models, offering reproducibility and transparency. However, their accessibility makes them likely targets of test set contamination. In this work, we introduce DyePack, a framework that leverages backdoor attacks to identify models that used benchmark test sets during training, without requiring access to the loss, logits, or any internal details of the model. Like how banks mix dye packs with their money to mark robbers, DyePack mixes backdoor samples with the test data to flag models that trained on it. We propose a principled design incorporating multiple backdoors with stochastic targets, enabling exact false positive rate (FPR) computation when flagging every model. This provably prevents false accusations while providing strong evidence for every detected case of contamination. We evaluate DyePack on five models across three datasets, covering both multiple-choice and open-ended generation tasks. For multiple-choice questions, it successfully detects all contaminated models with guaranteed FPRs as low as 0.000073% on MMLU-Pro and 0.000017% on Big-Bench-Hard using eight backdoors. For open-ended generation tasks, it generalizes well and identifies all contaminated models on Alpaca with a guaranteed false positive rate of just 0.127% using six backdoors.
Community
We present DyePack – a provably robust way to flag LLMs that train on benchmark test sets. No model loss or logits needed, and false positive rates are theoretically bounded and exactly computable.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- How Can I Publish My LLM Benchmark Without Giving the True Answers Away? (2025)
- STAMP Your Content: Proving Dataset Membership via Watermarked Rephrasings (2025)
- Wolf Hidden in Sheep's Conversations: Toward Harmless Data-Based Backdoor Attacks for Jailbreaking Large Language Models (2025)
- Detecting Stealthy Backdoor Samples based on Intra-class Distance for Large Language Models (2025)
- TrojanStego: Your Language Model Can Secretly Be A Steganographic Privacy Leaking Agent (2025)
- CoIn: Counting the Invisible Reasoning Tokens in Commercial Opaque LLM APIs (2025)
- Benign Samples Matter! Fine-tuning On Outlier Benign Samples Severely Breaks Safety (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper