Abdullah Garra (Abed)

Ph.D. Student · UMass Amherst

I am a first-year Ph.D. student in Computer Science at UMass Amherst, advised by Prof. Eugene Bagdasarian.

My research sits at the intersection of computer security and machine learning. I study how AI systems fail under adversarial pressure and how to make them trustworthy, from adversarial behavior and fingerprinting of deployed models, to secure LLM- and RAG-based systems, to using machine learning for security itself.

Before joining UMass Amherst, I completed my M.Sc. in Computer Science at Tel Aviv University, where I was a member of the Privacy, Learning, Usability, and Security (PLUS) research group advised by Dr. Mahmood Sharif.

Trustworthy AI AI Safety ML for Security Systems Security

Education

Sep 2026 – nowPh.D. in Computer Science, UMass Amherst
Advised by Prof. Eugene Bagdasarian
May 2024 – May 2026M.Sc. in Computer Science, Tel Aviv University
Advised by Dr. Mahmood Sharif · PLUS group
B.Sc. in Computer Science, Tel Aviv University

Selected Publications

arXiv
The TellTail of Embeddings: Fingerprinting Retrievers in Black-Box Systems

Abdullah Garra, Matan Ben-Tov, and Mahmood Sharif

arXiv, Oct 2026

Abs arXiv Bib PDF Code

Dense embedding models are core to modern text retrieval, enabling systems ranging from web search to retrieval-augmented generation (RAG). Yet, retrievers are usually deployed within opaque systems, exposing only ranked results, cited sources, or generated answers. This opacity prevents users from verifying which retrievers providers serve and may create a false sense of robustness against retrieval attacks. We show that attackers can infer retrievers’ identity only through queries. We introduce TellTail, a fingerprinting attack for identifying retrievers behind black-box systems across a spectrum of access levels. Despite sharing many properties, retrievers can be steered to emit distinct results. When retrieved passages are exposed, TellTail compares retrieval overlap to deduce the underlying model. When only the final generated response is exposed, TellTail optimizes model-specific queries that induce a chosen retrieval behavior on the target retriever but transfer poorly to others. Interestingly, the poor transferability that limits attacks elsewhere is exactly what makes them useful for fingerprinting. We evaluate TellTail using 53 retrievers under three increasingly restrictive settings. TellTail perfectly identifies the deployed retriever from full retrieval rankings; in 94.3% of attempts with unordered top-3 results; and in 92.5% of cases from language-model-generated answers alone. Furthermore, fingerprinting succeeds against a popular RAG system (OpenWebUI). Overall, TellTail shows that retrieval-based systems leak the identity of their embedding model—exposing intellectual property and enabling reconnaissance ahead of model-specific attacks such as corpus poisoning.
@article{garra2026telltail, title = {The TellTail of Embeddings: Fingerprinting Retrievers in Black-Box Systems}, author = {Garra, Abdullah and Ben-Tov, Matan and Sharif, Mahmood}, eprint = {2610.04026}, journal = {arXiv}, year = {2026}, month = oct, }
arXiv
Forecasting Cybersecurity Incidents Using Geopolitical Data and Large Language Models

Mark Fesenko^*, Abdullah Garra^*, Yaniv Harel, and 1 more author

arXiv, Oct 2026

Abs arXiv Bib PDF

Predicting security incidents is a profound task critical for informing proactive defensive measures and cyber-insurance policies. Prior work tackling this problem mainly utilized structured, manually defined features based on network measurements (e.g., protocol misconfigurations). Still, despite leading to promising performance, the network-based features may fail to capture aspects related to adversaries’ motives. To fill this gap, our work leverages geopolitical data mined from public sources–which may help capture attacker motives–to forecast security incidents. Specifically, our approach relies on news articles and transcribed podcasts that are fed to large language models to automatically produce rich representations. The representations are then fed to a classifier trained to forecast future incidents based on historical ones. Our evaluation with a large incidents dataset (>15,700 records) demonstrates substantial accuracy (71.4% ROC AUC) with geopolitical data alone. Notably, combining geopolitical data and network measurements outperforms the state-of-the-art technique based on network features alone (81.3% vs.77.5% ROC AUC). Our analysis also helps shed light on when geopolitical data is most helpful and the data sources that are most useful for accurate forecasting.
@article{fesenko2026forecasting, title = {Forecasting Cybersecurity Incidents Using Geopolitical Data and Large Language Models}, author = {Fesenko, Mark and Garra, Abdullah and Harel, Yaniv and Sharif, Mahmood}, eprint = {2610.04798}, journal = {arXiv}, year = {2026}, month = oct, }
arXiv
Towards Trustworthy Physical Intelligence: From Theory to Practice Across Life Cycle

Yang Wang, Hongxuan Liu, Xinghui Xu, and 40 more authors

Co-authors include Abdullah Garra and Yoshua Bengio

arXiv, Jul 2026

Abs arXiv Bib PDF

Physical intelligence refers to intelligence systems that understand, reason about, and act in accordance with the physical world and its underlying laws, dynamics, and constraints. Unlike conventional AI systems, physical intelligence interacts continuously with uncertain physical environments, and its actions produce consequences that are physically irreversible. As existing trustworthy AI frameworks have been developed primarily for digital AI systems, they do not fully capture the distinctive challenges of Physical Intelligence, such as physical safety, cyber-physical security, and physical manufacturing process. To address this gap, we present a survey of trustworthy physical intelligence principles. First, we characterize the core capabilities and challenges of physical intelligence. Second, we examine the role of physics in AI. Third, we trace the end-to-end physical intelligence life cycle across five core stages and introduce Trustworthy Physical Intelligence Operationalization (T-PAIO). Fourth, we develop the Trustworthy Physical Intelligence (T-PAI) framework, a theoretical framework that organizes key trustworthiness principles and provides a foundation for governing trustworthy physical intelligence systems.
@article{wang2026trustworthy, title = {Towards Trustworthy Physical Intelligence: From Theory to Practice Across Life Cycle}, author = {Wang, Yang and Liu, Hongxuan and Xu, Xinghui and Menon, Arjun and Cai, Xiaoran and He, Yunyu and Tarvo, Alex and Zhou, Jingzong and Ma, Mengzhong and Wei, Xinpeng and Yu, Yi and Wang, Shaobo and Peng, Cheng and Jiao, Aoran and Korolev, Alexei and Zhang, Yanyan and Ye, Kai and Li, Xinpeng and Guo, Chengquan and Fu, Jingjing and Bai, Nicholas and He, Yongjun and Ren, Junru and Ren, Silei and Shehab, Mohamad Louai and Cai, Keshu and Dennler, Nathaniel and Tus, Traian and Zhou, Gaoyue and Garra, Abdullah and Nakamura, Mason and Ortiz, George and Urbanas, Marius and Johannsmeier, Lars and Sharma, Rohit and Deshmukh, Suraj and Spranger, Michael and Gupta, Vinayak and Chen, Xu and Banerjee, Ashis G. and Feng, Yuxiang and Bengio, Yoshua and Qi, Peng}, eprint = {2607.22877}, journal = {arXiv}, year = {2026}, month = jul, }

News

Oct 05, 2026	🚀 Another paper is now on arXiv: Forecasting Cybersecurity Incidents Using Geopolitical Data and Large Language Models
Oct 05, 2026	🔥 Excited to be a co-author on Towards Trustworthy Physical Intelligence: From Theory to Practice Across Life Cycle, alongside an incredible group of researchers including Yoshua Bengio.
Oct 05, 2026	📄 TellTail, my first paper as lead author, is now on arXiv: The TellTail of Embeddings: Fingerprinting Retrievers in Black-Box Systems.
Sep 01, 2026	📄 Published my course project, Mitigating Emergent Collusion in LLM Pricing Agents, on arXiv for the community.
Jun 22, 2026	🧪 Acknowledged in TROPT: An Open Framework for Unifying and Advancing Discrete Text Optimization for assistance with experiments.

Projects & Datasets

Dataset · TellTail paper

msmarco-indexes

Pre-built FAISS indexes of the MS MARCO corpus for 53 embedding models, so retrieval experiments across many retrievers don't require re-indexing for each one.

View dataset →

Dataset · Forecasting paper

ccnews2024plus

A cleaned extension of the CC-News Common Crawl corpus beyond June 2024, continuing stanford-oval/ccnews, filtered and deduplicated with DataTrove.

View dataset →

Course project · ML with Graphs

advEGI

Does adversarial pretraining make transfer more stable? PGD-style adversarial training for Ego-Graph Information Maximization (EGI, NeurIPS 2021), evaluated across domains.

Read report →

Experience

Sep 2026 – nowGraduate Researcher, UMass Amherst
Dec 2023 – Oct 2026Graduate Researcher, Tel Aviv University

Awards

High School Excellence Scholarship for University Studies (merit-based; covered undergraduate tuition)

Outside the Lab

🏃 Sports & Outdoors

🏊‍♂️ Morning swims
🏋️‍♂️ Gym training
⚽ Soccer

🧠 Ideas & Thinking

🧩 Thought experiments
🎬 Debates & documentaries
⚖️ Ethics & philosophy

🎧 Music & Audio

🎶 Heavy music listener
🎙️ Long-form podcasts