VENOMREC: Cross-Modal Interactive Poisoning for Targeted Promotion in Multimodal LLM Recommender Systems

Guowei Guan, Yurong Hao, Jiaming Zhang, Tiantong Wu, Fuyao Zhang, Tianxiang Chen, Longtao Huang, Cyril Leung, Wei Yang Bryan Lim
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:37417-37433, 2026.

Abstract

Multimodal large language models (MLLMs) are pushing recommender systems (RecSys) toward content-grounded retrieval and ranking via cross-modal fusion. We find that while cross-modal consensus often mitigates conventional poisoning that manipulates interaction logs or perturbs a single modality, it also introduces a new attack surface where synchronised multimodal poisoning can reliably steer fused representations along stable semantic directions during fine-tuning. To characterise this threat, we formalise cross-modal interactive poisoning and propose VENOMREC, which performs Exposure Alignment to identify high-exposure regions in the joint embedding space and Cross-modal Interactive Perturbation to craft attention-guided coupled token–patch edits. Experiments on four real-world multimodal datasets demonstrate that VENOMREC consistently outperforms strong baselines, achieving 0.73 mean ER@20 and improving over the strongest baseline by +0.52 absolute ER points on average, while maintaining comparable recommendation utility. Code is available at https://github.com/GuoweiGuan666/VenomRec.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-guan26b, title = {{VENOMREC}: Cross-Modal Interactive Poisoning for Targeted Promotion in Multimodal {LLM} Recommender Systems}, author = {Guan, Guowei and Hao, Yurong and Zhang, Jiaming and Wu, Tiantong and Zhang, Fuyao and Chen, Tianxiang and Huang, Longtao and Leung, Cyril and Lim, Wei Yang Bryan}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {37417--37433}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/guan26b/guan26b.pdf}, url = {https://proceedings.mlr.press/v306/guan26b.html}, abstract = {Multimodal large language models (MLLMs) are pushing recommender systems (RecSys) toward content-grounded retrieval and ranking via cross-modal fusion. We find that while cross-modal consensus often mitigates conventional poisoning that manipulates interaction logs or perturbs a single modality, it also introduces a new attack surface where synchronised multimodal poisoning can reliably steer fused representations along stable semantic directions during fine-tuning. To characterise this threat, we formalise cross-modal interactive poisoning and propose VENOMREC, which performs Exposure Alignment to identify high-exposure regions in the joint embedding space and Cross-modal Interactive Perturbation to craft attention-guided coupled token–patch edits. Experiments on four real-world multimodal datasets demonstrate that VENOMREC consistently outperforms strong baselines, achieving 0.73 mean ER@20 and improving over the strongest baseline by +0.52 absolute ER points on average, while maintaining comparable recommendation utility. Code is available at https://github.com/GuoweiGuan666/VenomRec.} }
Endnote
%0 Conference Paper %T VENOMREC: Cross-Modal Interactive Poisoning for Targeted Promotion in Multimodal LLM Recommender Systems %A Guowei Guan %A Yurong Hao %A Jiaming Zhang %A Tiantong Wu %A Fuyao Zhang %A Tianxiang Chen %A Longtao Huang %A Cyril Leung %A Wei Yang Bryan Lim %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-guan26b %I PMLR %P 37417--37433 %U https://proceedings.mlr.press/v306/guan26b.html %V 306 %X Multimodal large language models (MLLMs) are pushing recommender systems (RecSys) toward content-grounded retrieval and ranking via cross-modal fusion. We find that while cross-modal consensus often mitigates conventional poisoning that manipulates interaction logs or perturbs a single modality, it also introduces a new attack surface where synchronised multimodal poisoning can reliably steer fused representations along stable semantic directions during fine-tuning. To characterise this threat, we formalise cross-modal interactive poisoning and propose VENOMREC, which performs Exposure Alignment to identify high-exposure regions in the joint embedding space and Cross-modal Interactive Perturbation to craft attention-guided coupled token–patch edits. Experiments on four real-world multimodal datasets demonstrate that VENOMREC consistently outperforms strong baselines, achieving 0.73 mean ER@20 and improving over the strongest baseline by +0.52 absolute ER points on average, while maintaining comparable recommendation utility. Code is available at https://github.com/GuoweiGuan666/VenomRec.
APA
Guan, G., Hao, Y., Zhang, J., Wu, T., Zhang, F., Chen, T., Huang, L., Leung, C. & Lim, W.Y.B.. (2026). VENOMREC: Cross-Modal Interactive Poisoning for Targeted Promotion in Multimodal LLM Recommender Systems. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:37417-37433 Available from https://proceedings.mlr.press/v306/guan26b.html.

Related Material