The paper examines how synthetic data can be used to covertly inject targeted social biases into aligned large language models. By having a misaligned teacher model generate filtered synthetic datasets in domains like creative writing and code generation, the authors fine‑tune aligned student models and demonstrate that the benign‑looking data can transmit biases while preserving general task performance. The study highlights a security risk in synthetic data‑driven LLM training pipelines and proposes log‑linearity‑based scoring as a potential safeguard.
By Minkyung Cho, Jihyo Kim, SeungWoo Song, Junghun Yuk, Minjoon Kee, Hoyun Song, KyungTae Lim
arXiv:2606. 31411v1 Announce Type: cross Abstract: Rapid advancements in generative speech technology have compromised the reliability of voice biometrics.
By Anh-Tuan Dao, Driss Matrouf, Mickael Rouvier, Nicholas Evans
Subliminal Learning (SL) is a surprising type of generalization displayed by modern language models. It allows the transfer of a bias or behavior from a teacher model to a student by distilling from seemingly unrelated or random synthetic data from the teacher.
arXiv:2608. 05734v1 Announce Type: new Abstract: Subliminal Learning (SL) is a surprising type of generalization displayed by modern language models.
By Ethan Hadley, Eren Gultepe
arXiv:2602. 04899v2 Announce Type: replace-cross Abstract: We present a data poisoning attack -- Phantom Transfer -- with the property that, even if you know precisely how the poison was placed into an otherwise benign dataset, you cannot filter it out.
By Andrew Draganov, Tolga H. Dur, Anandmayi Bhongade, Mary Phuong
arXiv:2606. 17464v1 Announce Type: new Abstract: Membership inference attacks (MIAs) are a canonical way to assess a machine learning model's privacy properties.
By Jeffrey G. Wang, Jason Wang, Marvin Li, Seth Neel
arXiv:2311.13892v4 Announce Type: replace-cross
Abstract: The social biases and unwelcome stereotypes revealed by pretrained language models are becoming obstacles to their application. Compared to n...
By Bingkang Shi, Xiaodan Zhang, Dehan Kong, Yulei Wu, Zongzhen Liu, Honglei Lyu, Longtao Huang
arXiv:2606. 28953v1 Announce Type: cross Abstract: Poisoning attacks entail attackers intentionally tampering with training data.
By Thomas Thebaud, Sonal Joshi, Henry Li, Martin Sustek, Jesus Villalba, Sanjeev Khudanpur, Najim Dehak
arXiv:2606. 10911v1 Announce Type: cross Abstract: Claims about the robustness and fairness of deepfake speech detectors are only as credible as the datasets used to train and evaluate those systems.
By Vojt\v{e}ch Stan\v{e}k, Eva Trnovsk\'a, Kamil Malinka, Anton Firc
arXiv:2607. 00325v1 Announce Type: new Abstract: A growing body of literature suggests that training data membership inference problems are fundamentally hard tasks in modern language modeling settings.
By John Kirchenbauer, Brian R. Bartoldson, Bhavya Kailkhura, Tom Goldstein
Backdoor attacks in Large Language Models (LLMs) are a growing security concern, where models can generate adversary-chosen content. Existing defenses target backdoors one at a time and typically require knowledge of the trigger, leaving the defender at a structural disadvantage when unknown backdoors may exist in a model.
arXiv:2610.07654v1 Announce Type: cross
Abstract: On-policy distillation (OPD) has attracted growing attention as an effective way to transfer capabilities from teacher models to student models. Rece...
By Jian Luo, Kehan Qi, Qingqiao Hu, Meilong Xu, Jiacheng Qiu, Weimin Lyu, Jiawei Zhou, Chao Chen