Hanxi Guo

About Me

I am currently a fourth-year Ph.D. student in Computer Science at Purdue University, advised by Prof. Xiangyu Zhang. Before joining Purdue, I received my M.Eng. and B.Eng. degrees in Computer Science from Shanghai Jiao Tong University (SJTU), where I was supervised by Prof. Ruhui Ma and Prof. Tao Song. I also worked with Prof. Yang Hua and Prof. Hao Wang.

My research aims to make AI systems more trustworthy and accountable by detecting the content they generate and auditing the actions they take. I currently focus on two directions:

AI Content Detection and Attribution: Developing methods to identify AI-generated content, trace its origins, and characterize human-AI collaboration.
AI Agent Auditing: Studying how to examine agent behaviors, identify potential failures, and improve the reliability and accountability of agentic workflows.

Previously, I worked on security and efficiency of distributed machine learning, particularly federated learning.

News

[Sept. 2026]Our paper “LensCT: Fine-Grained AI-Involved Text Detection via Temporal-Hierarchical Tomograms of LLM Internals.” is accepted by NeurIPS 2026!
[Aug. 2026]Our paper “Natural-Language Workflows Are Not Software Yet: Artifact-Driven Compilation for Reliable Agent Execution.” is released! We developed a workflow compiler that transforms natural-language workflows into executable, artifact-driven workflows through constrained optimization, improving the reliability of LLM-agent execution.
[May. 2026]Our paper “From Poisoned to Aware: Fostering Backdoor Self-Awareness in LLMs.” is accepted by ICML 2026 (Spotlight, Top 2.2%). Congrats to Guangyu!
[Nov. 2025]Our paper “Mitigating Backdoor Attacks via Trigger Reconstruction and Model Hardening.” is accepted by WACV 2026. Congrats to Guanhong!
[Nov. 2025]Our paper “Poisoning with A Pill: Circumventing Detection in Federated Learning.” is accepted by AAAI 2026!

[Nov. 2025]Four papers are accepted by NeurIPS 2025 Workshops on GenProCC, ResponsibleFM, and DL4C, respectively!

[Aug. 2025]Our paper “Profiler: Black-box AI-generated Text Origin Detection via Context-aware Inference Pattern Analysis.” is accepted by EMNLP 2025!

[Jul. 2025]Our team PurCL won the first place in the Amazon Nova AI Challenge! We have released our system description “ASTRA: Autonomous Spatial-Temporal Red-teaming for AI Software Assistants.”

[Jun. 2025]Our paper “JailbreakDiffBench: A Comprehensive Benchmark for Jailbreaking Diffusion Models.” is accepted by ICCV 2025. Congrats to Xiaolong!

[Jun. 2025]Our paper “SOFT: Selective Data Obfuscation for Protecting LLM Fine-tuning against Membership Inference Attacks.” is accepted by USENIX Security 2025. Congrats to Kaiyuan!

[Oct. 2024]Our paper “MultiVerse: Exposing Large Language Model Alignment Problems in Diverse Worlds.” is accepted by NeurIPS 2024 Workshop on Safe GenAI. Congrats to Xiaolong!

[Oct. 2024]Our paper “SkewAct: Red Teaming Large Language Models via Activation-Skewed Adversarial Prompt Optimization.” is accepted by NeurIPS 2024 Workshop on Red Teaming GenAI.

[Sept. 2024]Our paper “BiScope: AI-generated Text Detection by Checking Memorization of Preceding Tokens.” is accepted by NeurIPS 2024.

[Sept. 2024]Our paper “BAIT: Large Language Model Backdoor Scanning by Inverting Attack Target.” is accepted by S&P 2025. Congrats to Guangyu!

[Jul. 2024]Our new preprint “Poisoning with A Pill: Circumventing Detection in Federated Learning.” has been released, introducing a generic and attack-agnostic augmentation approach to enhance the effectiveness and stealthiness of existing poisoning attacks in federated learning.

[Jul. 2024]Our paper “UNIT: Backdoor Mitigation via Automated Neural Distribution Tightening.” is accepted by ECCV 2024. Congrats to Siyuan!

[Feb. 2024]Our paper “Our paper “Siren⁺: Robust Federated Learning with Proactive Alarming and Differential Privacy.” is accepted by IEEE TDSC.

[Aug. 2021]Our paper “Siren: Byzantine-robust Federated Learning via Proactive Alarming.” is accepted by SoCC 2021.

[Aug. 2021]Our paper “SpaceDML: Enabling Distributed Machine Learning in Space Information Networks.” is accepted by IEEE Network.

Featured Publications [Full List]

(* represents equal contribution)

NeurIPS

LensCT: Fine-Grained AI-Involved Text Detection via Temporal-Hierarchical Tomograms of LLM Internals

Hanxi Guo, Siyuan Cheng, Kaiyuan Zhang, Guangyu Shen, Xiangyu Zhang

Conference on Neural Information Processing Systems (NeurIPS), 2026.

Coming Soon

Preprint

Natural-Language Workflows Are Not Software Yet: Artifact-Driven Compilation for Reliable Agent Execution

Xiangzhe Xu^*, Hanxi Guo^*, Guangyu Shen, Siyuan Cheng, Xiangyu Zhang

arXiv preprint arXiv:2608.21341, 2026.

PDF

AAAI

Poisoning with A Pill: Circumventing Detection in Federated Learning

Hanxi Guo, Hao Wang, Tao Song, Tianhang Zheng, Yang Hua, Haibing Guan, Xiangyu Zhang

Annual AAAI Conference on Artificial Intelligence (AAAI), 2026.

PDF Code

NeurIPS Workshop

CodeMirage: A Multi-Lingual Benchmark for Detecting AI-Generated and Paraphrased Source Code from Production-Level LLMs

Hanxi Guo, Siyuan Cheng, Kaiyuan Zhang, Guangyu Shen, Xiangyu Zhang

NeurIPS 2025 Workshop on Deep Learning for Code in the Agentic Era (DL4C), 2025.

PDF Dataset

EMNLP

Profiler: Black-box AI-generated Text Origin Detection via Context-aware Inference Pattern Analysis

Hanxi Guo, Siyuan Cheng, Xiaolong Jin, Zhuo Zhang, Guangyu Shen, Kaiyuan Zhang, Shengwei An, Guanhong Tao, Xiangyu Zhang

Conference on Empirical Methods in Natural Language Processing (EMNLP), 2025.

PDF Code

NeurIPS

BiScope: AI-generated Text Detection by Checking Memorization of Preceding Tokens

Hanxi Guo, Siyuan Cheng, Xiaolong Jin, Zhuo Zhang, Kaiyuan Zhang, Guanhong Tao, Guangyu Shen, Xiangyu Zhang

Conference on Neural Information Processing Systems (NeurIPS), 2024.

PDF Code Dataset

NeurIPS Workshop

SkewAct: Red Teaming Large Language Models via Activation-Skewed Adversarial Prompt Optimization

Hanxi Guo, Siyuan Cheng, Guanhong Tao, Guangyu Shen, Zhuo Zhang, Shengwei An, Kaiyuan Zhang, Xiangyu Zhang

NeurIPS 2024 Workshop on Red Teaming GenAI: What Can We Learn from Adversaries?, 2024.

PDF

TDSC

Siren⁺: Robust Federated Learning with Proactive Alarming and Differential Privacy

Hanxi Guo, Hao Wang, Tao Song, Yang Hua, Ruhui Ma, Xiulang Jin, Zhengui Xue, Haibing Guan

IEEE Transactions on Dependable and Secure Computing (TDSC), 2024.

PDF Code

SoCC

Siren: Byzantine-robust Federated Learning via Proactive Alarming

Hanxi Guo, Hao Wang, Tao Song, Yang Hua, Zhangcheng Lv, Xiulang Jin, Zhengui Xue, Ruhui Ma, Haibing Guan

ACM Symposium on Cloud Computing (SoCC), 2021.

PDF Code

IEEE Network

SpaceDML: Enabling Distributed Machine Learning in Space Information Networks

Hanxi Guo, Qing Yang, Hao Wang, Yang Hua, Tao Song, Ruhui Ma, Haibing Guan

IEEE Network, 2021.

PDF