Changyi Li (李长艺) | Fudan University & Shanghai Innovation Institute

Changyi Li (李长艺)

Ph.D. Student

“I would devote 90% of my time to defining the right problem and the remaining 10% to solving it.”

Biography

My research focuses on safety, alignment, and capability development for frontier AI systems. I work on scalable oversight, risk evaluation, and recursive self-improvement in agentic systems, while also exploring embodied intelligence.

I'm always happy to discuss interesting ideas and potential collaborations. Feel free to reach out.

News

[09/2026]ReasoningShield was accepted to NeurIPS 2026.

[09/2026]Started the joint doctoral program between Fudan University and SII.

[07/2026]Invited to attend the OpenAI Safety & Alignment Mixer and the Seoul Alignment Workshop.

[05/2026]AutoControl Arena was accepted to ICML 2026.

Publications

* denotes equal contribution. For a complete publication record, please visit my Google Scholar profile.

AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluation

Changyi Li, Pengfei Lu, Xudong Pan, Fazl Barez, Min Yang

ICML 2026

[Page] [Paper] [Code] [Machine Heart] [PaperWeekly] [] 106

ReasoningShield: Safety Detection over Reasoning Traces of Large Reasoning Models

Changyi Li, Jiayi Wang, Xudong Pan, Geng Hong, Min Yang

NeurIPS 2026

[Page] [Paper] [Code] [Models] [Dataset] [] 26

ValueGraph overview: value orientations, pairwise priorities, and contextual evidence

ValueGraph: Evaluating and Representing Value Trade-offs in Large Language Models

Dylan Xinming Hou^*, Changyi Li^*, Chloe Ruyi Guo, Ziyuan Zhao, Gus Xia, Hua Shen

Preprint, 2026

[Page] [Paper] []

Gyges overview: controlled scenarios for testing hierarchical value stability

Gyges: Benchmarking Value Alignment as Hierarchical Value Stability

Dylan Xinming Hou^*, Changyi Li^*, Chloe Ruyi Guo, Ziyuan Zhao, Yichi Zhang, Hua Shen

Preprint, 2026

[Page] [Paper] []

CyberEvolver: Structured Self-Evolution for Cybersecurity Agents On the Fly

Yihe Fan, Changyi Li, Lichen Xu, Xudong Pan, Jiarun Dai, Hong Geng, Min Yang

Preprint, 2026

[Page] [Paper] [Code] [] 3

Large Language Model-Powered AI Systems Achieve Self-Replication with No Human Intervention

Xudong Pan, Jiarun Dai, Yihe Fan, Minyuan Luo, Changyi Li, Min Yang

Preprint, 2025

[Page] [Paper] [LiveScience] [Forbes] [The Independent] []

Honors

2026First-Class Scholarship, Fudan University
2025Freshman Scholarship, Fudan University
2024Principal Scholarship (1/38,000+)
2024Provincial Outstanding Graduate (Top 0.1%)
2023National Scholarship (Top 0.1%)
2023MCM/ICM Meritorious Winner, Team Leader
2022, 2023First-Class Scholarship & Merit Student Pacesetter

Academic Service

2026Invited Talk, Seoul Alignment Workshop, FAR.AI
2026Invited Participant, Safety & Alignment Closed-Door Mixer, OpenAI
2025Organizer, AI Deception Workshop, Fudan & SAIF & Concordia AI
2025, 2026Teaching Assistant, Foundation of Deep Learning, Fudan University