Yongli Xiang

Ph.D. Student @ University of Sydney

Hi, I am a first-year Ph.D. student at the University of Sydney, supervised by Prof. Tongliang Liu. I obtained my M.Sc. degree from USYD in 2025, also supervised by Prof. Liu. Before turning to academia, I worked as a Product Manager in Ads at Tencent. My research focuses on Generative AI Safety and Intellectual Property (IP) Protection for AI models, aiming to make advanced AI systems safer in what they generate and more controllable in how they generalize.

Scholar Code Dataset Email LinkedIn

News

2026.06📢 Our ECCV LifeGenIP Workshop call for papers and challenge is now open.
2026.06🎤 Invited by NVIDIA to give a Tech Talk at CVPR 2026. Thanks for the invitation!
2026.03🎉 One paper (When Safety Collides) accepted to CVPR 2026.
2025.12🏅 Named in the USYD Engineering Postgraduate High Honour Roll. Thanks for the recognition!
2025.07✨ Received OpenAI Researcher Access Program support. Thanks for the support!

Selected Publications

* Equal contribution; † Corresponding author

When Agents Learn to Be You: Benchmarking Privacy Leakage, Impersonation Risk, and Defenses in Persona Skills
Yongli Xiang*, Zhifang Zhang*, Bojun Yang, Ziming Hong, Lei Feng, Miao Xu, Tongliang Liu†.
arXiv preprint, 2026 GenAI Safety
We introduce AntiSkillBench, an end-to-end benchmark for measuring privacy leakage, impersonation risk, and defense effectiveness across the persona-skill pipeline.
When Safety Collides: Resolving Multi-Category Harmful Conflicts in Text-to-Image Diffusion via Adaptive Safety Guidance
Yongli Xiang, Ziming Hong†, Zhaoqing Wang, Xiangyu Zhao, Bo Han, Tongliang Liu†.
CVPR, 2026 GenAI Safety
We propose CASG, a training-free framework that resolves conflicts among harmful categories by adaptively steering text-to-image diffusion toward the most relevant safety direction.
Jailbreaking the Non-Transferable Barrier via Test-Time Data Disguising.
Yongli Xiang, Ziming Hong†, Lina Yao, Dadong Wang, Tongliang Liu†.
CVPR, 2025 NTL for IP Protection
We introduce JailNTL, a black-box attack that bypasses non-transferable learning defenses by disguising unauthorized data at test time without modifying model weights.
Toward Robust Non-Transferable Learning: A Survey and Benchmark
Ziming Hong*, Yongli Xiang*, Tongliang Liu†.
IJCAI, Survey Track, 2025 NTL for IP Protection
We present the first comprehensive survey of non-transferable learning and NTLBench, a unified benchmark that reveals the robustness limitations of existing methods.
VII: Visual Instruction Injection for Jailbreaking Image-to-Video Generation Models
Bowen Zheng, Yongli Xiang, Ziming Hong, Zerong Lin, Chaojian Yu, Tongliang Liu, Xinge You.
arXiv Preprint, 2026 GenAI Safety
We uncover visual instruction injection, showing how benign-looking reference images can carry hidden malicious intent to jailbreak commercial image-to-video models.

Academic Services

LifeGenIP: Life-Cycle Intellectual Property Governance of Visual Generative Models
Organizer, ECCV Workshop, 2026 IP Protection
We organize LifeGenIP, a forum on IP governance across the full life cycle of visual generative models, from data collection to deployment.

Teaching

2026 S2Teaching Assistant: COMP5328 Advanced Machine Learning
2025 S1Tutor: COMP5046 Natural Language Processing