DIS-CO: Discovering Copyrighted Content in VLMs Training Data

André V. Duarte, Xuandong Zhao, Arlindo L. Oliveira, Lei Li
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:14807-14832, 2025.

Abstract

How can we verify whether copyrighted content was used to train a large vision-language model (VLM) without direct access to its training data? Motivated by the hypothesis that a VLM is able to recognize images from its training corpus, we propose DIS-CO, a novel approach to infer the inclusion of copyrighted content during the model’s development. By repeatedly querying a VLM with specific frames from targeted copyrighted material, DIS-CO extracts the content’s identity through free-form text completions. To assess its effectiveness, we introduce MovieTection, a benchmark comprising 14,000 frames paired with detailed captions, drawn from films released both before and after a model’s training cutoff. Our results show that DIS-CO significantly improves detection performance, nearly doubling the average AUC of the best prior method on models with logits available. Our findings also highlight a broader concern: all tested models appear to have been exposed to some extent to copyrighted content. We provide the code in the supplementary materials.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-duarte25a, title = {{DIS}-{CO}: Discovering Copyrighted Content in {VLM}s Training Data}, author = {Duarte, Andr\'{e} V. and Zhao, Xuandong and Oliveira, Arlindo L. and Li, Lei}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {14807--14832}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/duarte25a/duarte25a.pdf}, url = {https://proceedings.mlr.press/v267/duarte25a.html}, abstract = {How can we verify whether copyrighted content was used to train a large vision-language model (VLM) without direct access to its training data? Motivated by the hypothesis that a VLM is able to recognize images from its training corpus, we propose DIS-CO, a novel approach to infer the inclusion of copyrighted content during the model’s development. By repeatedly querying a VLM with specific frames from targeted copyrighted material, DIS-CO extracts the content’s identity through free-form text completions. To assess its effectiveness, we introduce MovieTection, a benchmark comprising 14,000 frames paired with detailed captions, drawn from films released both before and after a model’s training cutoff. Our results show that DIS-CO significantly improves detection performance, nearly doubling the average AUC of the best prior method on models with logits available. Our findings also highlight a broader concern: all tested models appear to have been exposed to some extent to copyrighted content. We provide the code in the supplementary materials.} }
Endnote
%0 Conference Paper %T DIS-CO: Discovering Copyrighted Content in VLMs Training Data %A André V. Duarte %A Xuandong Zhao %A Arlindo L. Oliveira %A Lei Li %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-duarte25a %I PMLR %P 14807--14832 %U https://proceedings.mlr.press/v267/duarte25a.html %V 267 %X How can we verify whether copyrighted content was used to train a large vision-language model (VLM) without direct access to its training data? Motivated by the hypothesis that a VLM is able to recognize images from its training corpus, we propose DIS-CO, a novel approach to infer the inclusion of copyrighted content during the model’s development. By repeatedly querying a VLM with specific frames from targeted copyrighted material, DIS-CO extracts the content’s identity through free-form text completions. To assess its effectiveness, we introduce MovieTection, a benchmark comprising 14,000 frames paired with detailed captions, drawn from films released both before and after a model’s training cutoff. Our results show that DIS-CO significantly improves detection performance, nearly doubling the average AUC of the best prior method on models with logits available. Our findings also highlight a broader concern: all tested models appear to have been exposed to some extent to copyrighted content. We provide the code in the supplementary materials.
APA
Duarte, A.V., Zhao, X., Oliveira, A.L. & Li, L.. (2025). DIS-CO: Discovering Copyrighted Content in VLMs Training Data. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:14807-14832 Available from https://proceedings.mlr.press/v267/duarte25a.html.

Related Material