I work on the alignment and preparedness of LLMs and AI agents, with the goal of making increasingly capable systems safe by design. Over the past three years, my work has focused on LLM post-training and evaluation, for both safety and capability. My earlier research studied the dynamics of agent behavior in AI systems.

Papers

Please see my Google Scholar profile for a full list of publications.

$^\dagger$ equal contribution

[1] D-FUSEr: Diverse Failure, Unified Success via Error-Distribution Shaping in LLM Reasoning. Andrew Estornell $^\dagger$, David D. Baek $^\dagger$, Yichi Zhang, Muhammad Faaiz Taufiq, Jean-Francois Ton, Jie Mei, Tao Wang. International Conference on Machine Learning (ICML) 2026.

[2] Evian: Towards Explainable Visual Instruction-tuning Data Auditing. Zimu Jia, Mingjie Xu, Andrew Estornell, Jiaheng Wei. Association for Computational Linguistics (ACL) 2026.

[3] Any-Depth Alignment: Unlocking Innate Safety Alignment of LLMs to Any-Depth. Jiawei Zhang, Andrew Estornell, Bo Li, David D. Baek, Xiaojun Xu. International Conference on Learning Representations (ICLR) 2026.

[4] How to Train a Leader: Hierarchical Reasoning in Multi-Agent LLMs. Andrew Estornell $^\dagger$, Jean-Francois Ton $^\dagger$, Muhammad Faaiz Taufiq $^\dagger$, Hang Li. Preprint 2025.

[5] ACC-Collab: An Actor Critic Approach to Multi-Agent LLM Collaboration. Andrew Estornell $^\dagger$, Jean-Francois Ton $^\dagger$, Yuanshun Yao, Yang Liu. International Conference on Learning Representations (ICLR) 2025. [paper], [pdf]

[6] AgenticMath: Enhancing LLM Reasoning via Agentic-based Math Data Generation. Xianyang Liu, Yilin Liu, Shuai Wang, Hao Cheng, Andrew Estornell, Yuzhi Zhao, Jun Shu, Jiaheng Wei. Preprint 2025.

[7] Multi-LLM Debate: Framework, Principals, and Interventions. Andrew Estornell, Yang Liu. Advances in Neural Information Processing Systems (NeurIPS) 2024. [paper]

[8] Measuring and Reducing LLM Hallucination Without Gold-Standard Answers via Expertise-Weighting. Jiaheng Wei, Yuanshun Yao, Jean-Francois Ton, Hongyi Guo, Andrew Estornell, Yang Liu. Preprint 2024. [paper]

[9] User-Creator Feature Polarization in Recommender Systems with Dual Influence. Tao Lin, Kun Jin, Andrew Estornell, Xiaoying Zhang, Yiling Chen, Yang Liu. Advances in Neural Information Processing Systems (NeurIPS) 2024. [paper]

[10] Which Features are the Fairest of them All? The Impact of Features Used by Algorithms on Perceptions of Fairness. Andrew Estornell $^\dagger$, Tina Zhang $^\dagger$, Sanmay Das, Chien-Ju Ho, Brendan Juba, Yevgeniy Vorobeychik. International Joint Conference on Artificial Intelligence (IJCAI) 2024.

[11] Adaptive Recruitment Resource Allocation to Improve Cohort Representativeness in Participatory Biomedical Datasets. Victor Borza, Andrew Estornell, Ellen Wright Clayton, Chien-Ju Ho, Russell Rothman, Yevgeniy Vorobeychik, Bradley Malin. American Medical Informatics Association (AMIA) 2024 (awarded best student paper).

[12] Dataset Representativeness and Downstream Task Fairness. Victor Borza, Andrew Estornell, Chien-Ju Ho, Bradley Malin, Yevgeniy Vorobeychik. Preprint 2024.

[13] Incentivizing Recourse through Auditing in Strategic Classification. Andrew Estornell, Sanmay Das, Yang Liu, Yatong Chen, Yevgeniy Vorobeychik. International Joint Conference on Artificial Intelligence (IJCAI) 2023.

[14] Unfairness Despite Awareness: Group-Fair Classification with Strategic Agents. Andrew Estornell, Sanmay Das, Yang Liu, Yevgeniy Vorobeychik. Conference on Fairness, Accountability, and Transparency (FAccT) 2023. Also appeared in Learning with Strategic Agents (LSA) 2022 (awarded best paper). [link] [pdf]

[15] Popularizing Fairness: Group Fairness and Individual Welfare. Andrew Estornell, Sanmay Das, Brendan Juba, Yevgeniy Vorobeychik. Conference on Artificial Intelligence (AAAI) 2023. [link] [pdf]

[16] Location Spoofing Attacks on Autonomous Fleets. Jinghan Yang, Andrew Estornell, Yevgeniy Vorobeychik. Conference on Vehicle Security and Privacy (VehicleSec) 2023. [paper]

[17] Predicting Customer Goals in Financial Institution Services: A Data-Driven LSTM Approach. Andrew Estornell $^\dagger$, Stylianos Loukas Vasileiou $^\dagger$, William Yeoh, Daniel Borrajo, Rui Silva. ICAPS Workshop on Planning for Financial Services (FinPlan) 2023. [paper]

[18] Manipulating Elections by Changing Voter Perceptions. Junlin Wu, Andrew Estornell, Lecheng Kong, Yevgeniy Vorobeychik. International Joint Conference on Artificial Intelligence (IJCAI) 2022. [link] [pdf]

[19] Incentivizing Truthfulness Through Audits in Strategic Classification. Andrew Estornell, Sanmay Das, Yevgeniy Vorobeychik. Conference on Artificial Intelligence (AAAI) 2021. [link] [pdf]

[20] Election Control by Manipulating Issue Significance. Andrew Estornell, Sanmay Das, Edith Elkind, Yevgeniy Vorobeychik. Conference on Uncertainty in Artificial Intelligence (UAI) 2020. [link] [pdf]

[21] Deception Through Half-Truths. Andrew Estornell, Sanmay Das, Yevgeniy Vorobeychik. Conference on Artificial Intelligence (AAAI) 2020. [link] [pdf]

[22] PBW Deformations of Quadratic Monomial Algebras. Andrew Estornell $^\dagger$, Zachary Cline $^\dagger$, Chelsea Walton $^\dagger$, Matthew Wynne $^\dagger$. Communications in Algebra 2019. [link] [pdf]