Hi! I am currently a Postdoctoral Researcher under joint supervision between the Institute of Automation, Chinese Academy of Sciences (CASIA) and Wenge AI Lab. I was affilated in State Key Laboratory of Multimodal Artificial Intelligence Systems. I received my PhD from CASIA in 2024 and my bachelor’s degree from Southwest University in 2019.

My research interests include:

• Alignment of LLM/VLM (RL, interpretablity)
• Safety Assessment of LLM/VLM (Jailbreak, red-teaming)
• Adversarial Robustness (mainly during my Phd period)

News

• Aug. 2025: The helpful-only model released on Hugging Face has surpassed 35k downloads.
• July 2025: Awarded 2nd place in the ICML 2025 AI4MATH Challenge (Track 2: SeePhys).

Publications

• Jiahao Zhao, Liwei Dong. Jinx: Unlimited LLMs for Probing Alignment Failures. technical report, 2025.

• Jiahao Zhao, Wenji Mao, Daniel Zeng. Disentangled Text Representation Learning with Information-Theoretic Perspective for Adversarial Robustness. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2024.

• Yilin Cao, Jiahao Zhao, Rui Zhang, Hanyi Zou, Wenji Mao. TARA: Token-level Attribute Relation Adaptation for Multi-Attribute Controllable Text Generation. Findings of the Association for Computational Linguistics, EMNLP 2024.

• Xingjin Wang, Jiahao Zhao, Jiahui Shi, Linjing Li, Daniel Zeng. A Novel Visual-Enhanced Dual Stream Long-Term Decision Framework for Large Language Model Agents. International Conference on Neural Information Processing, ICONIP 2024.

• Minzheng Wang, Nan Xu, Jiahao Zhao, Yin Luo, Wenji Mao. PromISe: Releasing the Capabilities of LLMs with Prompt Introspective Search. Proceedings of the 2024 Joint International Conference on Computational Linguistics, LREC-COLING 2024.

• Jiahao Zhao, Minzheng Wang, Nan Xu, Yin Luo, Wenji Mao. Enhancing Adversarial Robustness of LLMs with Analytic Hierarchy Process. First Conference on Language Modeling, COLM 2024.

• Jiahao Zhao, Wenji Mao. Generative Adversarial Training with Perturbed Token Detection for Model Robustness. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023.

• Jiahao Zhao, Penghui Wei, Wenji Mao. Robust Neural Text Classification and Entailment via Mixup Regularized Adversarial Training. Proceedings of the 44th International ACM SIGIR Conference, SIGIR 2021.

• Penghui Wei, Jiahao Zhao, Wenji Mao. A Graph-to-Sequence Learning Framework for Summarizing Opinionated Texts. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021.

• Penghui Wei, Jiahao Zhao, Wenji Mao. Effective Inter-Clause Modeling for End-to-End Emotion-Cause Pair Extraction. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, ACL 2020.

Experience

• Jan. 2025 – Present. Postdoc, CASIA & Wenge. Mainly working on AI4S LLM/Agent development.

• Nov. 2023 – Dec. 2024. Intern, Luminlp @ miHoYo Inc. Lived a happy time here. Got a bitter lesson as well.

  说到做到，有话直说，只认功劳，追求极致

• June. 2023 – Sep. 2023. Intern, AI Lab @ Wenge. Witnessed the first wave of LLMs.

Awards

• 2024: “Climbing” Second-Class Scholarship, Institute of Automation, CASIA
• 2023: Outstanding Student Award, Institute of Automation, CASIA
• 2019: Outstanding Graduate, Southwest University
• 2018: National Scholarship, Ministry of Education