|
I am a fourth year CS Ph.D. candidate at University of Illinois, Urbana-Champaign (UIUC), advised by Prof. Bo Li. I received my Bachelor's degree from CS, Zhejiang University at CKC Honors College, advised by Prof. Shouling Ji and Prof. Siliang Tang. My research focuses on making LLMs, VLMs, and AI agents safe, scalable, and interpretable at real-world scale. I work on scalable red-teaming to uncover realistic agent failures, interpretability-guided analysis of reasoning and hallucinations, and efficient post-training for robust deployment. My long-term goal is to build AI systems that are not only powerful but also reliable and practically deployable across real-world applications. Email / Google Scholar / Github / LinkedIn |
|
News
| 2025/09 - One paper got accepted to NeurIPS 2025. |
| 2025/05 - I started my internship at Meta, working on VLM reasoning and interpretability. |
| 2025/05 - One paper got accepted to ICML 2025. |
| 2025/04 - We released UltraLong-8B, models with up to 4M context length and competitive performance on standard tasks. |
| 2025/01 - Four papers got accepted to ICLR 2025. |
| 2024/12 - Two papers got accepted to AAAI 2025. |
| 2024/09 - We released ChatQA 2, a Llama 3.0-based model with enhanced long-context understanding and RAG capabilities. |
| 2024/09 - Our paper, DecodingTrust, got the Cybersecurity award 2024 on Best Machine Learning and Security Paper. |
| 2024/05 - I started my internship at NVIDIA, working on long context LLMs. |
| 2024/04 - We are hosting the The Competition for LLM and Agent Safety 2024! |
| 2024/02 - One paper got accepted to CVPR 2024. |
| 2023/12 - Our paper, DecodingTrust, received the Outstanding Paper award at NeurIPS 2023. |
Selected Publications
Full publication list available on my Google Scholar.
|
GuardSet-X: Massive Multi-Domain Safety Policy-Grounded Guardrail Dataset Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS), 2025 [PDF] [Code] [BibTeX] |
|
AdvAgent: Controllable Blackbox Red-teaming on Web Agents Chejian Xu,Forty-Second International Conference on Machine Learning (ICML), 2025 [PDF] [Code] [Website] [BibTeX] |
|
From 128K to 4M: Efficient Training of Ultra-Long Context Large Language Models Chejian Xu,Preprint, 2025 [PDF] [Website] [Model Weights 🤗] [BibTeX] |
|
MMDT: Decoding the Trustworthiness and Safety of Multimodal Foundation Models Chejian Xu,The Thirteenth International Conference on Learning Representations (ICLR), 2025 [PDF] [Code] [Website] [T2I Dataset 🤗] [I2T Dataset 🤗] [BibTeX] |
|
EIA: Environmental Injection Attack on Generalist Web Agents for Privacy Leakage The Thirteenth International Conference on Learning Representations (ICLR), 2025 [PDF] [Code] [BibTeX] |
|
ChatQA 2: Bridging the Gap to Proprietary LLMs in Long Context and RAG Capabilities Peng Xu,The Thirteenth International Conference on Learning Representations (ICLR), 2025 [PDF] [Website] [Model Weights 🤗] [Training Data 🤗] [BibTeX] |
|
AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models The Thirteenth International Conference on Learning Representations (ICLR), 2025 [PDF] [BibTeX] |
|
DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS), 2023 (Outstanding Paper) [PDF] [Code] [Website] [BibTeX] |
|
SafeBench: A Benchmarking Platform for Safety Evaluation of Autonomous Vehicles Chejian Xu*,Thirty-sixth Conference on Neural Information Processing Systems (NeurIPS), 2022 [PDF] [Code] [Leaderboard] [BibTeX] |
|
SemAttack: Natural Textual Attacks via Different Semantic Spaces North American Chapter of the Association for Computational Linguistics (NAACL), 2022 (Findings) [PDF] [Code] [BibTeX] |
|
Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS), 2021 (Oral) [PDF] [Leaderboard] [Dataset] [BibTeX] |
Service
| Conference Reviewer: NeurIPS 2022-2025, ICML 2025, ICLR 2025-2026, CVPR 2026, ICCV 2025, ACL 2025, EMNLP 2025, AISTATS 2025-2026, AAAI 2023-2025 |
| Organizer: The Competition for LLM and Agent Safety 2024, CVPR 2023 SSAD Workshop, NeurIPS 2022 DMLW Workshop |
| Program Committee: ICLR 2025 SynthData Workshop, ICLR 2023 RTML Workshop |