Hi there Welcome to my Homepage!

Hi! My name is Xutao Mao. I am a first-year Ph.D. student at the City University of Hong Kong, advised by Prof. Cong Wang. I also closely work with Prof. Xiang Zheng and Prof. Bo Han. Before that, I received my B.S. with Honors in Computer Science and Mathematics from Vanderbilt University.

My research asks how AI fails as it grows more capable, and how to keep that evolution safe. As AI starts to act in the world, it also begins to coordinate with others, then keep memory. I study the safety surface that appears at each of those steps, with a focus on AI safety. Feel free to reach out if you are interested in collaboration.

Research Interests

Acting and coordinating. I study how AI fails while it acts in the world and while it hands work to other AI.
Remembering and improving. I study how AI failure persists in memory and later changes what it does.
Looking inside and stepping in. I study the internal signals behind those failures and how to intervene from inside the model.

News

2026.09 🚀🚀 We released work on identity hijacking in agent orchestration and on agents breaking correct work when falsely accused on arXiv.
2026.09 🎉🎉 Our work on mechanistic analysis and mitigation for CoT obfuscation in VLMs was accepted to NeurIPS 2026.
2026.09 🎓🎓 I am starting my Ph.D. at the City University of Hong Kong.
2026.08 🎉🎉 Our work on watermarking for agent long-term memory was accepted to Findings of EMNLP 2026.
2026.07 🚀🚀 We released work on autoresearch to discovery vulnerability in production agent on arXiv.
2026.05 🎉🎉 Our work on red-teaming for multi-modal toxicity was accepted to ICML 2026 (Poster).
2025.12 🎓🎓 I received my B.S. in Computer Science with Honors from Vanderbilt University.
2025.11 🎉🎉 Our work on social-media opinion prediction (Oral) and on multi-domain text-to-SQL benchmark (Poster) accepted to AAAI 2026.

Experience

City University of Hong Kong
2026.09 - Present
Ph.D. in Computer Science, advised by Prof. Cong Wang
Research: AI safety.

Hong Kong Baptist University
2025.12 - 2026.08
Research Assistant, advised by Prof. Bo Han
Research: CoT monitorability.

Vanderbilt University
2022.08 - 2025.12
B.S. with Honors in Computer Science & Mathematics

Publications

(* equal contribution · † corresponding author)

NeurIPS'26

Taming CoT Obfuscation in VLMs: From Mechanistic Evidence to Activation-Level Enforcement Xutao Mao*, Jianing Zhu*, Jinman Zhao, Tongliang Liu, Xiaowen Chu, Cong Wang†, Bo Han† Advances in Neural Information Processing Systems, 2026

PAPER CODE

ICML'26

STARE: Step-wise Temporal Alignment and Red-teaming Engine for Multi-modal Toxicity Attack Xutao Mao, Liangjie Zhao, Tao Liu, Xiang Zheng†, Hongying Zan, Cong Wang† Proceedings of the 43rd International Conference on Machine Learning, 2026

PAPER CODE

Preprint

Trust the Brand, Lose Control: How Identity Hijacks LLM Agent Orchestration Xutao Mao*, Rui Qian*, Linghan Chen, Yudong Gao, Junchi Liao, Jiulin Cai, Jinman Zhao, Cong Wang† Preprint, 2026

PAPER CODE PROJECT DATASET

Preprint

"You're Right, Let Me Fix It": How LLM Agents Damage Correct Work When Falsely Accused Xutao Mao*, Rui Qian*, Longxiang Wang, Xinjian Yi, Mingxuan Li, Linghan Chen, Yudong Gao, Xiang Zheng†, Cong Wang† Preprint, 2026

PAPER CODE PROJECT DATASET

Preprint

Agent Hacks Agents: Autoresearch Discovers Vulnerabilities in Production Agents Xutao Mao, Rui Qian, Xiang Zheng†, Cong Wang† Preprint, 2026

PAPER CODE PROJECT

NeurIPS 2026 Taming CoT Obfuscation in VLMs: From Mechanistic Evidence to Activation-Level Enforcement
Xutao Mao*, Jianing Zhu*, Jinman Zhao, Tongliang Liu, Xiaowen Chu, Cong Wang†, Bo Han† [paper][code]
EMNLP 2026 Findings MemMark: State-Evolution Attribution Watermarking for Agent Long-Term Memory Systems
Haobo Zhang*, Xutao Mao*, Guangyuan Dong, Ziwei Li†, Xuanbo Su, Kaijie Chen, Jing Yang, Zheng Lin [paper][code][project]
ICML 2026 Poster STARE: Step-wise Temporal Alignment and Red-teaming Engine for Multi-modal Toxicity Attack
Xutao Mao, Liangjie Zhao, Tao Liu, Xiang Zheng†, Hongying Zan, Cong Wang† [paper][code]
AAAI 2026 Oral MindVote: When AI Meets the Wild West of Social Media Opinion
Xutao Mao†, Ezra Xuanru Tao, Leyao Wang [paper][code]
AAAI 2026 Poster LogicCat: Text-to-SQL Benchmark for Multi-Domain Reasoning Challenges
Tao Liu*, Xutao Mao*, Hongying Zan†, Dixuan Zhang, Yifan Li, Haixin Liu, Lulu Kong, Jiaming Hou, Rui Li, YunLong Li, Aoze Zheng, Zhiqiang Zhang, Zhewei Luo, Kunli Zhang, Min Peng * equal contribution, listed in alphabetical order [paper][code]
Preprint Trust the Brand, Lose Control: How Identity Hijacks LLM Agent Orchestration
Xutao Mao*, Rui Qian*, Linghan Chen, Yudong Gao, Junchi Liao, Jiulin Cai, Jinman Zhao, Cong Wang† [paper][code][project][dataset]
Preprint "You're Right, Let Me Fix It": How LLM Agents Damage Correct Work When Falsely Accused
Xutao Mao*, Rui Qian*, Longxiang Wang, Xinjian Yi, Mingxuan Li, Linghan Chen, Yudong Gao, Xiang Zheng†, Cong Wang† [paper][code][project][dataset]
Preprint Agent Hacks Agents: Autoresearch Discovers Vulnerabilities in Production Agents
Xutao Mao, Rui Qian, Xiang Zheng†, Cong Wang† [paper][code][project]
Preprint Agents Don't Just Agree, They Remember: Benchmarking Persistent Sycophancy in Self-Improving Personal Agents
Xutao Mao*, Liangjie Zhao*, Leyao Wang, Rui Qian, Qiang Huang, Wentao Wang, Bo Han†, Xiang Zheng†, Cong Wang† [paper][code][project][dataset]
Preprint What Happens Inside Agent Memory? Circuit Analysis from Emergence to Diagnosis
Xutao Mao, Jinman Zhao, Gerald Penn, Cong Wang† [paper]
Preprint Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents
Xutao Mao, Liangjie Zhao, Xiang Zheng†, Cong Wang† [paper][code]
Preprint Towards Bridging Review Sparsity in Recommendation with Textual Edge Graph Representation
Leyao Wang*, Xutao Mao*, Xuhui Zhan, Yuying Zhao, Bo Ni, Ryan A Rossi, Nesreen K Ahmed, Tyler Derr† [paper]
Preprint ParlAI Vote: A Web Platform for Analyzing Gender and Political Bias in Large Language Models
Wenjie Lin*, Hange Liu*, Yingying Zhuang, Xutao Mao, Jingwei Shi, Xudong Han, Tianyu Shi, Jinrui Yang† [paper]
Preprint Cooldown Landmines: Cross-Tenant Interference Attacks on LLM Gateways
Yudong Gao*, Linghan Chen*, Wenhan Wu, Quan Shi, Xutao Mao, Mia Zhou, Junjian Li, Xiaolong Liu, Jiyao Wang, Mingyu Guo, Honglong Chen† [paper]

Scholarships

2026, Hong Kong Postgraduate Scholarships (during Ph.D. study).
2025, Vanderbilt Summer Research Program Scholarship.

Services

Reviewer, ICLR 2027 · NeurIPS 2026 · AAAI 2026 / 2027 · TheWebConf 2026 · ACL ARR (Oct 2026).

Collaboration

Let's discuss ideas and build something meaningful together.

I'm always happy to discuss new research ideas and explore potential collaborations, especially around AI safety as AI grows more capable. If our interests overlap, feel free to reach out.

xutao.henry.mao@gmail.com