My name is Jialin Wu. I graduated from the USSLAB at Zhejiang University with a master’s degree in Electrical Engineering, where I worked on AI security under the guidance of Prof. Wenyuan Xu and Prof. Yanjiao Chen.
My research primarily focuses on AI security and privacy, with a specific emphasis on the security and safety of multimodal large language models (MLLMs).
I joined Ant Group in July 2025, where I continue to work on AI safety and security.
🔥 News
- 2026.05: 🎉 Revis was accepted by ICML 2026!
- 2025.09: 🎉 EnchTable was accepted by IEEE S&P 2026!
- 2025.06: 🎉 Graduated and joined Ant Group.
- 2024.12: ✨ Started an internship at Ant CPLab.
- 2024.10: 🌍 Attended CCS 2024 in Salt Lake City—my first international trip!
- 2024.08: 🎉 Legilimens was accepted by CCS 2024! Feeling so lucky.
📝 Publications
Revis: Sparse Latent Steering to Mitigate Object Hallucination in Large Vision-Language Models
Jialin Wu, Wei Shi, Han Shen, Peigui Qi, Kunsheng Tang, Zhicong Huang, Binghao Wang, Zhou Yang.
ICML 2026 (CCF-A) [Code] Acceptance rate: 26.6%
Revis is a sparse latent steering framework for mitigating object hallucination in large vision-language models.
EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models
Jialin Wu, Kecen Li, Zhicong Huang, Xinfeng Li, Xiaofeng Wang, Cheng Hong.
IEEE S&P 2026 (CCF-A, Big4) [Code] [Model] Acceptance rate: 12.75%
EnchTable is a framework designed to transfer safety alignment to fine-tuned downstream models. EnchTable effectively preserves model utility while significantly improving safety across diverse architectures and task domains.
Legilimens: Practical and Unified Content Moderation for Large Language Model Services
Jialin Wu, Jiangyi Deng, Shengyuan Pang, Yanjiao Chen, Jiayang Xu, Xinfeng Li, Wenyuan Xu.
ACM CCS 2024 (CCF-A, Big4) [Code] Acceptance rate: 16.7%
Legilimens is a practical and unified content moderation framework that achieves effective and efficient moderation by extracting conceptual features from chat-oriented LLMs, despite their conversational fine-tuning.
PATRONUS: Safeguarding Text-to-Image Models against White-Box Adversaries
Xinfeng Li, Shengyuan Pang, Jialin Wu, Jiangyi Deng, Huanlong Zhong, Yanjiao Chen, Jie Zhang, Wenyuan Xu
🎖 Honors and Awards
- 2024–2025 Excellent Graduate (honor for graduation), Zhejiang University.
- 2023–2024 🌟 National Scholarship, Zhejiang University.
- 2023–2024 Outstanding Graduate Student, Zhejiang University.
📚 Services
Journal Reviewer: ACM Transactions on Privacy and Security
📖 Educations
- 2022.09 - 2025.06, Zhejiang University, Master in Electrical Engineering.
💻 Internships
- 2024.12 - 2025.06, Crypto&Privacy lab, Ant Research, Ant Group, Hangzhou, Zhejiang, China.
- 2024.06 - 2024.10, Trust and Safety, AIGC Safety, ByteDance, Hangzhou, Zhejiang, China.