Yuxuan Zhou

Yuxuan Zhou

Trustworthy generative AI · RL post-training

Hello! I’m Yuxuan Zhou (周愈轩), a second-year Ph.D. student at the Shenzhen International Graduate School, Tsinghua University, advised by Professor Shu-Tao Xia and Professor Tao Dai. I received my B.Eng. in Cybersecurity (Qiming Honor Class) from Huazhong University of Science and Technology in 2025.

My research focuses on the safety and alignment of foundation models, and the detection of AI-generated content. I believe both will matter for a long time to come. Reinforcement learning runs through both lines — as the way to align models, and as the way to train detectors that generalize.

I have worked on these problems at ByteDance and Tencent Hunyuan. I’m always happy to talk about research, collaborations, or internship opportunities — feel free to reach out.

Google Scholar  ·  GitHub  ·  Email

🔬 Research

Safety & Alignment of Foundation Models. As models gain tools and autonomy, the cost of misalignment grows with them. I probe where alignment fails — through automated red-teaming and jailbreaking of multimodal models — and repair it with RL-based post-training that treats safety and capability as a joint objective rather than a trade-off.

AI-Generated Content Detection. Detectors that hold up against unseen forgeries in the real world. To close the benchmark-to-deployment gap, we use generative models to synthesize forgery types and corruptions missing from public data, so detectors learn artifacts that transfer rather than dataset-specific shortcuts. I lead a team building a deployed detection platform for the anti-fraud program of the Yunnan Provincial Public Security Department.

🔥 News

  • 2026.08:  🎉 Four papers were accepted by EMNLP 2026 (2 Main, 2 Findings), including one in the Main Conference as first author!
  • 2026.06:  🎖 I joined the Foundation Model Department at Tencent Hunyuan as a research intern, working on foundation model safety and alignment.
  • 2025.12:  🎉 One paper on AI-generated image detection was accepted by ICASSP 2026!
  • 2025.11:  🎉 One paper (as first author) on deepfake detection was accepted by AAAI 2026!
  • 2025.06:  🎊 I received my B.Eng. from Huazhong University of Science and Technology (HUST) and started my Ph.D. in August 2025. Thank you to all my advisors and collaborators!
  • 2025.03:  🎉 One paper on backdoor sample detection was accepted by IEEE S&P 2025!
  • 2025.03:  🎖 I joined Flow Security at ByteDance as a research intern, mentored by Yang Bai!
  • 2024.11:  🎉 One paper on adversarial examples was accepted by AAAI 2025 as an Oral!
  • 2024.07:  🥇 Our work won Special Prize (only 12 nationwide, first in HUST history) and First Prize in the National College Student Information Security Competition.

📝 Publications

* denotes first author. Full list available on Google Scholar.

Safety & Alignment of Foundation Models

AI-Generated Content Detection

Others

💻 Experience

  • 2026.06 - 2026.08, Tencent Hunyuan, Foundation Model Department, China — foundation model safety and alignment.
  • 2025 - 2026, ByteDance, Flow Security, China — mentored by Yang Bai.

📖 Education

  • 2025.08 - Now, Ph.D. in Computer Science, Shenzhen International Graduate School, Tsinghua University, Shenzhen, China.
  • 2021.09 - 2025.06, B.Eng. in Cybersecurity (Qiming Honor Class), Huazhong University of Science and Technology, Wuhan, China.  GPA 3.92/4.00, rank 2/18.

🎖 Honors and Awards

  • 2024.07: The 17th National College Student Information Security Competition — Special Prize (12 nationwide) and First Prize. First in HUST history.
  • 2024.03: Qiushi Cup, Silver Award (only one in the department).
  • 2023.07: National-level Undergraduate Innovation & Entrepreneurship Training Program (only one in the department).
  • 2022.01: Shenyang Aircraft Corporation “Flying Shark” Scholarship.

🛠 Academic Service

  • Reviewer: AAAI 2026, AAAI 2027, ACL 2026.