Arman Zharmagambetov Research Scientist Meta Superintelligence Labs (FAIR) Menlo Park, California, USA Email: armanz [at] meta [dot] com

About

I am a Research Scientist in the Meta's Superintelligence Labs (FAIR) . My research primarily focuses on machine learning and optimization, with recent projects exploring their application in enhancing the security and robustness of AI models. Before that, I was appointed as a Postdoctoral Researcher at FAIR, where I worked with Yuandong Tian on AI-guided optimization, reinforcement learning and combinatorial optimization.

I completed my Ph.D. at University of California, Merced (UCM) advised by Miguel Á. Carreira-Perpiñán, where I primarily studied learning algorithms for decision trees and tree-based models (see TAO).

CV and biography

Selected Publications

[See all publications] [Google Scholar ]

• [ICML] A. Paulus*, I. Kulikov, B. Amos, R. Munos, I. Evtimov, K. Chaudhuri, A. Zharmagambetov*: "Safety Alignment of LMs via Non-cooperative Games".
  (* = Equal technical contribution)
  [external link] [arXiv] [code]

• S. Chen*, A. Zharmagambetov, D. Wagner, C. Guo*: "Meta SecAlign: A Secure Foundation LLM Against Prompt Injection Attacks".
  (* = Equal technical contribution)
  [arXiv] [code] [Model: 8B] [Model: 70B]

• [ICML] A. Zharmagambetov*, A. Paulus*, C. Guo, B. Amos**, and Y. Tian**: "AdvPrompter: Fast Adaptive Adversarial Prompting for LLMs".
  (* = Equal 1st authors, ** = Equal advising)
  International Conference on Machine Learning, 2025.
  [external link] [arXiv] [code]

• [NeurIPS] A. Zharmagambetov, C. Guo*, I. Evtimov*, M. Pavlova, R. Salakhutdinov and K. Chaudhuri: "AgentDAM: Privacy Leakage Evaluation for Autonomous Web Agents".
  (* = Equal contribution)
  Advances in Neural Information Processing Systems (Datasets & Benchmarks Track) , 2025.
  [arXiv] [code]

• [NeurIPS] A. Zharmagambetov*, I. Evtimov*, A. Grattafiori, C. Guo** and K. Chaudhuri**: "WASP: Benchmarking Web Agent Security Against Prompt Injection Attacks".
  (* = Equal 1st authors, ** = Equal advising)
  Advances in Neural Information Processing Systems (Datasets & Benchmarks Track) , 2025.
  [arXiv] [code]

• [NeurIPS] A. Zharmagambetov, B. Amos, A. Ferber, T. Huang, B. Dilkina, and Y. Tian: "Landscape Surrogate: Learning Decision Losses for Mathematical Optimization Under Partial Information".
  Advances in Neural Information Processing Systems, 2023.
  [arXiv] [code]

• [NeurIPS] A. Zharmagambetov and M. Á. Carreira-Perpiñán: "Semi-Supervised Learning with Decision Trees: Graph Laplacian Tree Alternating Optimization".
  Advances in Neural Information Processing Systems, 2022.
  [external link] [paper preprint] [short video] [poster]

• [ICML] A. Zharmagambetov and M. Á. Carreira-Perpiñán: "Smaller, More Accurate Regression Forests Using Tree Alternating Optimization".
  International Conference on Machine Learning, 2020.
  [external link] [paper preprint] [supplementary material] [slides] [video]

Professional Activities

• Reviewer, Journal of Machine Learning Research (JMLR), since 2023.
• Reviewer, Int. Conference on Learning Representations (ICLR), since 2021 (Outstanding Reviewer).
• Reviewer, Conference on Neural Information Processing Systems (NeurIPS), since 2019 (Best Reviewer).
• Reviewer, Int. Conference on Machine Learning (ICML), since 2020.
• Reviewer, Int. Conference on Artificial Intelligence and Statistics (AISTATS 2022).
• Reviewer, AAAI Conf. on Artificial Intelligence (AAAI): 2020, 2021.

Media Coverage

• Invited guest @Zuckerberg.korshileri [in Kazakh]: Link
• Invited guest @PhD_talks [in Kazakh]: Link
• Invited guest @Narikbi_LIVE [in Kazakh]: Part-1, Part-2
• Invited guest @nFactorial Podcast [in Russian]: Link
• Article in Forbes Kazakhstan [in Russian]: Link

FAIR | EECS, UC Merced | My Home Page