AI security & privacy
Training-data extraction, privacy auditing, and the security of code language models.
Codebreaker · IEEE S&P 2025 ↗About
Hi there.
It’s Changzhou Han.
PhD Researcher Swinburne University of Technology
My research focuses on privacy and security in AI systems, particularly code language models and autonomous agents. I also work on sign language translation and multimodal learning.
Before starting my PhD, I studied computer science at Swinburne and information security at Xidian University.
01 / Research
Research interests
AI agent security & safety
Threat analysis for autonomous agents and safety dynamics in human–AI interaction.
Multimodal learning
Long-form sign language translation, temporal segmentation, and diffusion-based semantic reconstruction.
BoostSLT · CVPR 2026 ↗02 / Selected publications
Selected publications
CVPR 2026Conference paper
BoostSLT: Boosting Sign Language Translation via a Plug-and-Play Diffusion-Based Semantic Enhancer
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 28828–28837.
A model-agnostic framework combining temporal segmentation and diffusion-based semantic reconstruction for long-form sign language translation.
IEEE S&P 2025Conference paper
Codebreaker: Dynamic Extraction Attacks on Code Language Models
2025 IEEE Symposium on Security and Privacy (S&P), pp. 559–575.
An automated extraction framework for studying personal-information leakage from the training data of code language models.
03 / Publications
All publications
Loading publications…
Publication details checked against Google Scholar and publisher records, October 2026. Preprints are labeled separately.
04 / Background
Background
Doctor of Philosophy
Swinburne University of Technology Australia
Bachelor of Computer Science (Honours)
Swinburne University of Technology Australia
Security Researcher
Qihoo360 · SkyGo Team
Bachelor’s Degree in Information Security
Xidian University China
05 / Contact
Contact
You can reach me by email for questions about my work
or potential collaborations.