👋🏻 Welcome!

Hello! I am Zhihao Liu (刘至昊), a Ph.D. student in Cyberspace Security at Zhejiang University (ZJU). I started my Ph.D. studies in 2023. I work on trustworthy and privacy-preserving artificial intelligence, with a particular focus on the security, privacy, and reliability of large language models and AI agents.

My research spans LLM agent security, prompt injection defenses, retrieval-augmented generation security, machine unlearning, differential privacy, and privacy risks in modern language and vision-language models.

📬 I am open to research discussions and collaborations. Please feel free to contact me if our interests overlap.

🔎 Research Interests

  • LLM and Agent Security: Understanding and mitigating prompt injection, adversarial manipulation, and other security risks in tool-using language-model agents.
  • Privacy-Preserving Machine Learning: Developing practical differential privacy methods and studying privacy leakage in large language and vision-language models.
  • Machine Unlearning: Designing and evaluating methods that remove the influence of selected training data while preserving model utility.
  • Trustworthy Retrieval and Inference: Improving the security and reliability of retrieval-augmented generation, long-context inference, and model-serving systems.

Publications

For a complete and current list, please visit my Google Scholar profile.

2026

  • [NeurIPS] Z. Liu, Z. Guo, S. Shao, Y. He, W. Wang, M. Chen, D. Wang, and Y. Hu, “Ask in the Crowd: Differentially Private LLM Inference via Dummy-Augmented Shuffling,” NeurIPS'26. [Paper]
  • [USENIX Sec] Y. He, H. Zhu, Y. Li, S. Shao, H. Yao, Z. Liu, and Z. Qin, “AttriGuard: Defeating Indirect Prompt Injection in LLM Agents via Causal Attribution of Tool Invocations,” USENIX Sec'26. [Paper]
  • [NDSS] Z. Luo, S. Shao, S. Zhang, L. Zhou, Y. Hu, C. Zhao, Z. Liu, and Z. Qin, “Shadow in the Cache: Unveiling and Mitigating Privacy Risks of KV-Cache in LLM Inference,” NDSS'26. [Paper]
  • [TIFS] Z. Liu, J. Lou, W. Bao, Y. Hu, W. Wenli, Z. Qin, and K. Ren, “Differentially Private Zeroth-Order Methods for Scalable Large Language Model Finetuning,” IEEE TIFS, 2026. [Paper]
  • [TDSC] Z. Liu, J. Lou, Y. Hu, X. Li, Y. Cheng, T. Chen, Z. Qin, Z. Qin, and K. Ren, “Towards Mitigating Excessive Forgetting in LLM Unlearning via Entanglement-Guidance with Proxy Constraint,” IEEE TDSC, 2026. [Paper]
  • [ACM CCS] T. Chen, Y. He, Y. Wang, S. Shao, H. Zheng, Z. Liu, J. Li, Z. Qin, Y. Chen, Z. Chu, et al., “MIRAGE: Misleading Retrieval-Augmented Generation via Black-Box and Query-Agnostic Poisoning Attacks,” ACM CCS'26. [Paper]
  • [arXiv] Z. Liu, Y. Wu, J. Lou, D. Wang, Y. Zhou, and Y. Hu, “Aligned but Fragile: Enhancing LLM Safety Robustness via Zeroth-Order Optimization,” arXiv'26. [Paper]
  • [arXiv] Z. Liu, Y. Cheng, S. Liu, R. Wen, Y. Hu, and Z. Qin, “DP-MIND: Differentially Private Memory-Guided Long Context Inference,” arXiv'26. [Paper]

2025

  • [USENIX Sec] Y. Hu, Z. Li, Z. Liu, Y. Zhang, Z. Qin, K. Ren, and C. Chen, “Membership Inference Attacks against Vision-Language Models,” USENIX Sec'25. [Paper]
  • [arXiv] W. Bao, J. Lou, Y. Hu, X. Li, Z. Liu, J. Liu, Z. Qin, and K. Ren, “Module-Aware Parameter-Efficient Machine Unlearning on Transformers,” arXiv'25. [Paper]
  • [arXiv] K. Miao, Y. Hu, X. Li, W. Bao, Z. Liu, Z. Qin, and K. Ren, “Towards Evaluation for Real-World LLM Unlearning,” arXiv'25. [Paper]

📖 Education

2023.09 – Present

Ph.D. in Cyberspace Security

Zhejiang University, Hangzhou, China

2019.09 – 2023.06

B.Eng. in Cyberspace Security

Shandong University, Qingdao, China