avatar

Mohamed Aghzal
محمد أغزال
ⵎⵓⵃⴰⵎⴷ ⴰⵖⵣⴰⵍ

Ph.D. Student George Mason University

maghzal (at) gmu (dot) edu

About Me

I am a Ph.D. student at George Mason University, where I have the fortune of being co-advised by Dr. Ziyu Yao and Dr. Gregory J. Stein. I also collaborate frequently with Dr. Erion Plaku. I am a member of the George Mason NLP Group. Prior to that I was part of the Darija NLP Lab at Al Akhawayn University, where I worked on low-resource languages and dialectical Arabic.

My research focuses on understanding and improving how intelligent agents reason, plan, and act in complex environments, with particular emphasis on evaluating, interpreting, and controlling agent behavior. I study why LLM agents fail, how their decision-making processes give rise to these failures, and how we can intervene to improve their behavior in both physical (e.g., robotics) and digital environments (e.g., web navigation).

Ultimately, my goal is to understand the inner working of LLM-based Agents in order to make them are safer, more reliable and controllable.

Research Interests

Planning and Reasoning: Planning and reasoning are crucial components of human intelligence, enabling us to navigate complex environments, make informed decisions, and achieve our goals. A central research inquiry in my work involves exploring ways to leverage foundation models to build systems that can reason and plan.
Explainability and Interpretability: I am interested in understanding the mechanisms underlying agent behavior, bridging behavioral analysis and mechanistic interpretability to explain how internal computations give rise to decisions across multi-step interactions.
Agent Controllability and Safety: I investigate how to reliably steer agent behavior through targeted interventions, with particular interest in developing methods and evaluation frameworks that connect mechanistic understanding to effective and generalizable control.

News

  • October 2026 Check out our new position paper on LLM agent interpretability!
  • August 2026 Our team (GMU-Veritaserum - Antik Ayan Khan, Mohamed Aghzal, and Prof. Ziyu Yao) won best white box approach at the Aletheia's Quest LLM lie detection challenge!
  • July 2026 Heading to San Diego ☀️ for ACL'26
  • May 2026 Check out our new paper on mechanistic interpretability for LLM agents
  • May 2026 Received a Distinguished Academic Award from the Computer Science Department at GMU 🏆
  • May 2026 Attended the North East AI Agents Day in New York City 🗽
  • April 2026 Paper accepted to ACL 2026 🎉
  • March 2026 Check out our new preprint on hierarchical evaluation of LLM Web Agents
  • June 2025 Attended IEEE/CVF CVPR 2025 in Nashville, TN 🎸
  • April 2025 Received a Travel Award for IEEE/CVF CVPR 2025
  • Feb 2025 Attended AAAI 2025 in Philadelphia, PA
  • Feb 2025 Paper accepted to IEEE/CVF CVPR 2025
  • Feb 2025 Check out our new survey on the use of LLMs in Planning
  • Nov 2024 Check out our new preprint
  • June 2024 Paper accepted to IEEE CASE 2024
  • Mar 2024 Paper accepted to LLM Agents at ICLR 2024
  • Oct 2023 New preprint available on ArXiv
  • Apr 2023 Presented a poster at MASC-SLL 2023

Publications

  1. Preprint
    Understanding LLM Agents Requires a Chain of Attribution: On the Interplay Between Mechanistic Interpretability and Behavioral Approaches
    Mohamed Aghzal, Mor Geva, Wajih Ul Hassan, Gregory J. Stein, Ziyu Yao
    Preprint.
  2. Preprint
    Toward Mechanistic Interpretability of LLM Agents: Explaining Trajectories Demands New Methods
    Mohamed Aghzal, Mor Geva, Gregory J. Stein, Ziyu Yao
    Preprint.
  3. ACL 2026
    Why Do LLM-based Web Agents Fail? A Hierarchical Planning Perspective
    Mohamed Aghzal, Gregory J. Stein, Ziyu Yao
    The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026)
  4. ArXiv
    A Survey on Large Language Models for Automated Planning
    Mohamed Aghzal, Erion Plaku, Gregory J. Stein, Ziyu Yao
    ArXiv:2502.12435
  5. CVPR 2025
    Evaluating Vision-Language Models as Evaluators in Path Planning
    Mohamed Aghzal, Xiang Yue, Erion Plaku, Ziyu Yao
    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2025)
  6. CASE 2024
    Look Further Ahead: Testing the Limits of GPT-4 in Path Planning
    Mohamed Aghzal, Erion Plaku, Ziyu Yao
    IEEE 20th International Conference on Automation Science and Engineering
  7. LLMAgents@ICLR
    Can Large Language Models be Good Path Planners? A Benchmark and Investigation on Spatial-temporal Reasoning
    Mohamed Aghzal, Erion Plaku, Ziyu Yao
    ICLR Workshop on LLM Agents, 2024
  8. Cogn. Comput. 2024
    Investigating Offensive Language Detection in a Low-Resource Setting with a Robustness Perspective
    Israe Abdellaoui, Anass Ibrahimi, Mohamed Amine El Bouni, Asmaa Mourhir, Saad Driouech, Mohamed Aghzal
    Big Data and Cognitive Computing
  9. MASC-SLL 2023
    English to Linear Temporal Logic Translation using Neural Language Models [Poster]
    Mohamed Aghzal, Ziyu Yao, Erion Plaku
    The 10th Mid-Atlantic Student Colloquium on Speech, Language and Learning (MASC-SLL 2023)
  10. CiST
    Compact Transformer-based Language Models for the Moroccan Darija
    Mohamed Aghzal, Mohamed Amine El Bouni, Saad Driouech, Asmaa Mourhir
    7th IEEE Congress on Information Science and Technology (CiST)
  11. ACLing 2021
    Distributional Word Representations for Code-Mixed Text in Moroccan Darija
    Mohamed Aghzal, Asmaa Mourhir
    Procedia Computer Science
  12. ICDS 2020
    Early Diagnosis of Parkinson's Disease Based on Handwritten Patterns Using Deep Learning
    Mohamed Aghzal, Asmaa Mourhir
    IEEE Fourth International Conference on Intelligent Computing in Data Sciences (ICDS 2020)

Education

Ph.D. in Computer Science — George Mason University, Fairfax, VA, USA
M.Sc. in Computer Science — George Mason University, Fairfax, VA, USA
B.Sc. in Computer Science — Al Akhawayn University, Ifrane, Morocco


Experience

01/2023 - : Graduate Research Assistant at George Mason University, Fairfax, VA, USA
08/2022 - : Graduate Teaching Assistant at George Mason University, Fairfax, VA, USA
09/2021 - 08/2022: Adjunct Faculty Member at Al Akhawayn University, Ifrane, Morocco
06/2021 - 08/2021: Research Intern at Attijariwafa Bank, Casablanca, Morocco
06/2020 - 08/2022: Research Assistant at Al Akhawayn University in Ifrane, Morocco


Teaching

CS 310 - Data Structures: Graduate Teaching Assistant (Fall 2023, Spring 2026)
CS 262 - Introduction to Low Level Programming: Lab Instructor (Fall 2022, Spring 2023)
CSC 3326 L - Database Systems: Instructor (Fall 2021)
CSC 1401 L - Computer Programming: Instructor (Fall 2021, Spring 2022, Summer 2022)
CSC 2303 - Object-Oriented Programming: Group Tutor (Fall 2019, Spring 2021)
CSC 4399 - Deep Learning: Teaching Assistant (Fall 2018, Fall 2020)
CSC 2302 - Data Structures: Grader (Spring 2019)
CSC 1400 - Introduction to Computers: Grader (Spring 2019)
CSC 1401 - Computer Programming: Grader (Fall 2018)
CSC 2303 - Object-Oriented Programming: Grader (Fall 2018)

Services

Reviewer: IEEE CiST 2023, IEEE CASE 2024, EMNLP 2024 , IROS 2025 , ACL 2025

Secondary Reviewer: NeurIPS 2023, CoLM 2025