Md Jahangir Alam

Md Jahangir Alam

LLM & Agent Security, Privacy, Applied ML

Sitemap

A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.

Pages

Posts

Notes, soon

less than 1 minute read

Published:

Placeholder post. I plan to write here about agent security evaluation — what the benchmarks we build actually measure, where tool-use guardrails give false assurance, and the parts of this work that do not fit in a paper.

portfolio

publications

Automl Systems for Medical Imaging

Published in Book chapter, Springer, 2023

Summary placeholder — to be updated.

Recommended citation: Tasmia Tahmida Jidney, Angona Biswas, Nasim Md Abdullah Al, Ismail Hossain, Md Jahangir Alam, Sajedul Talukder, Mofazzal Hossain, Md Azim Ullah (2023). "Automl Systems for Medical Imaging." Book chapter, Springer.
Download Paper

Introduction to Medical Imaging Informatics

Published in Book chapter, Springer, 2023

Medical imaging informatics is a rapidly growing field that combines the principles of medical imaging and informatics to improve the acquisition, management, and interpretation of medical images.

Recommended citation: Md Zihad Bin Jahangir, Ruksat Hossain, Riadul Islam, Nasim Md Abdullah Al, Md Mahim Anjum Haque, Md Jahangir Alam, Sajedul Talukder (2023). "Introduction to Medical Imaging Informatics." Book chapter, Springer.
Download Paper

Online Learning for X-Ray, CT or MRI

Published in Book chapter, Springer, 2023

Summary placeholder — to be updated.

Recommended citation: Mosabbir Bhuiyan, Nasim Md Abdullah Al, Sarwar Saif, Kishor Datta Gupta, Md Jahangir Alam, Sajedul Talukder (2023). "Online Learning for X-Ray, CT or MRI." Book chapter, Springer.
Download Paper

Combating Identity Attacks in Online Social Networks: A Multi-Layered Framework Using Zero-Knowledge Proof and Permissioned Blockchain

Published in ASONAM 2023, 2023

Identity attacks, such as impersonation, identity theft, and fraudulent account creation, pose significant threats to the security and trustworthiness of Online Social Networks (OSNs).

Recommended citation: Md Jahangir Alam, Ismail Hossain, Sai Puppala, Sajedul Talukder (2023). "Combating Identity Attacks in Online Social Networks: A Multi-Layered Framework Using Zero-Knowledge Proof and Permissioned Blockchain." ASONAM 2023.
Download Paper

Monitoring Dynamics of Emotional Sentiment in Social Network Commentaries

Published in ASONAM 2023, 2023

The proliferation of social media offers a real-time reflection of public sentiments. Sentiment analysis on such platforms yields crucial insights for sectors like market research, politics, business strategy, and public health.

Recommended citation: Ismail Hossain, Sai Puppala, Md Jahangir Alam, Sajedul Talukder (2023). "Monitoring Dynamics of Emotional Sentiment in Social Network Commentaries." ASONAM 2023.
Download Paper

Towards Addressing Identity Deception in Social Media using Bangla Text-Based Gender Identification

Published in ASONAM 2023, 2023

Gender identification from social media content can play a crucial role in detecting and mitigating the risks posed by counterfeit accounts. Authentic gender representation can foster a safer and more diverse online environment.

Recommended citation: Sultan Ahmed, Md Jahangir Alam, Sajedul Talukder, Ismail Hossain (2023). "Towards Addressing Identity Deception in Social Media using Bangla Text-Based Gender Identification." ASONAM 2023.
Download Paper

Blockchain-Integrated Secure Framework for Enhanced E-Government Services

Published in Book chapter, Springer (Signals and Communication Technology), 2023

Summary placeholder — to be updated.

Recommended citation: Sajedul Talukder, Md Jahangir Alam, Ismail Hossain, Sai Puppala (2023). "Blockchain-Integrated Secure Framework for Enhanced E-Government Services." Book chapter, Springer (Signals and Communication Technology).
Download Paper

Advancements in Multimodal Social Media Post Summarization: Integrating GPT-4 for Enhanced Understanding

Published in COMPSAC 2024, 2024

The proliferation of social media platforms, such as Facebook, has led to an exponential increase in diverse user-generated content, including text, images, and videos that could be pivotal in public health contexts.

Recommended citation: Md Jahangir Alam, Ismail Hossain, Sai Puppala, Sajedul Talukder (2024). "Advancements in Multimodal Social Media Post Summarization: Integrating GPT-4 for Enhanced Understanding." COMPSAC 2024.
Download Paper

SCALE: Self-Regulated Clustered FederAted LEarning in a Homogeneous Environment

Published in COMPSAC 2024, 2024

Federated Learning (FL) has emerged as a transfor-mative approach for enabling distributed machine learning while preserving user privacy, yet it faces challenges like communication inefficiencies and reliance on centralized infrastructures, leading to increased latency and costs.

Recommended citation: Sai Puppala, Ismail Hossain, Md Jahangir Alam, Zahidur Talukder, Syed Bahauddin, Sajedul Talukder (2024). "SCALE: Self-Regulated Clustered FederAted LEarning in a Homogeneous Environment." COMPSAC 2024.
Download Paper

SCAN: A HealthCare Personalized ChatBot with Federated Learning Based GPT

Published in COMPSAC 2024, 2024

Our paper introduces a groundbreaking approach to healthcare information retrieval and engagement through a personalized chatbot system empowered by Federated Learning Based GPT.

Recommended citation: Sai Puppala, Ismail Hossain, Md Jahangir Alam, Sajedul Talukder (2024). "SCAN: A HealthCare Personalized ChatBot with Federated Learning Based GPT." COMPSAC 2024.
Download Paper

SocialGuard: Bangla Text-Based Gender Identification for Enhancing Integrity in Social Networks

Published in COMPSAC 2024, 2024

In this study, we address the task of discerning gender through the textual content of social media, a crucial step in detecting and mitigating counterfeit account activity. Ensuring accurate gender portrayal on digital platforms is essential for creating a secure and inclusive cyberspace.

Recommended citation: Md Jahangir Alam, Sultan Ahmed, Ismail Hossain, Sai Puppala, Zahidur Talukder, Sajedul Talukder (2024). "SocialGuard: Bangla Text-Based Gender Identification for Enhancing Integrity in Social Networks." COMPSAC 2024.
Download Paper

Generative AI like ChatGPT in Blockchain Federated Learning: use cases, opportunities and future

Published in Preprint (arXiv), 2024

Federated learning has become a significant approach for training machine learning models using decentralized data without necessitating the sharing of this data.

Recommended citation: Sai Puppala, Ismail Hossain, Md Jahangir Alam, Sajedul Talukder, Jannatul Ferdaus, Mahedi Hasan, Sameera Pisupati, Shanmukh Mathukumilli (2024). "Generative AI like ChatGPT in Blockchain Federated Learning: use cases, opportunities and future." Preprint (arXiv).
Download Paper

A visual approach to tracking emotional sentiment dynamics in social network commentaries

Published in ICWSM 2024 · Social Network Analysis and Mining, 2024

The expansion of social media has unlocked a real-time barometer of public opinion. This paper introduces a novel framework to analyze sentiment shifts in social network comment sections, a reflection of the broader public discourse over time.

Recommended citation: Ismail Hossain, Sai Puppala, Md. Jahangir Alam, Sajedul Talukder (2024). "A visual approach to tracking emotional sentiment dynamics in social network commentaries." ICWSM 2024 · Social Network Analysis and Mining.
Download Paper

Optimus-Q: Utilizing Federated Learning in Adaptive Robots for Intelligent Advanced Reactor Operations Through Quantum Cryptography

Published in NPIC&HMIT 2025, 2025

Summary placeholder — to be updated.

Recommended citation: Sai Puppala, Ismail Hossain, Jahangir Alam, Kazuma Kobayashi, Samrendra Roy, Syed Alam, Sajedul Talukder (2025). "Optimus-Q: Utilizing Federated Learning in Adaptive Robots for Intelligent Advanced Reactor Operations Through Quantum Cryptography." NPIC&HMIT 2025.
Download Paper

EVOLVE-X: Embedding Fusion and Language Prompting for User Evolution Forecasting on Social Media

Published in Preprint (arXiv), 2025

Social media platforms serve as a significant medium for sharing personal emotions, daily activities, and various life events, ensuring individuals stay informed about the latest developments.

Recommended citation: Ismail Hossain, Sai Puppala, Md Jahangir Alam, Sajedul Talukder (2025). "EVOLVE-X: Embedding Fusion and Language Prompting for User Evolution Forecasting on Social Media." Preprint (arXiv).
Download Paper

Comprehensive Privacy Risk Assessment in Social Networks Using User Attributes Social Graphs and Text Analysis

Published in HyperText 2025, 2025

The rise of social networking platforms has amplified privacy threats as users increasingly share sensitive information across profiles, content, and social connections.

Recommended citation: Md Jahangir Alam, Ismail Hossain, Sai Puppala, Sajedul Talukder (2025). "Comprehensive Privacy Risk Assessment in Social Networks Using User Attributes Social Graphs and Text Analysis." HyperText 2025.
Download Paper

Variational Gaussian Mixture Manifold Models for Client-Specific Federated Personalization

Published in Preprint (arXiv), 2025

Personalized federated learning (PFL) often fails under label skew and non-stationarity because a single global parameterization ignores client-specific geometry.

Recommended citation: Sai Puppala, Ismail Hossain, Md Jahangir Alam, Sajedul Talukder (2025). "Variational Gaussian Mixture Manifold Models for Client-Specific Federated Personalization." Preprint (arXiv).
Download Paper

LLM-Guided Dynamic-UMAP for Personalized Federated Graph Learning

Published in Preprint (arXiv), 2025

We propose a method that uses large language models to assist graph machine learning under personalization and privacy constraints.

Recommended citation: Sai Puppala, Ismail Hossain, Md Jahangir Alam, Tanzim Ahad, Sajedul Talukder (2025). "LLM-Guided Dynamic-UMAP for Personalized Federated Graph Learning." Preprint (arXiv).
Download Paper

A Comprehensive Survey of Federated Learning for Edge AI: Recent Trends and Future Directions

Published in Preprint, 2025

Federated learning (FL) has emerged as a key enabler of privacy-preserving, distributed intelligence, yet most deployments and surveys remain cloud-centric.

Recommended citation: Sai Puppala, Ismail Hossain, Jahangir Alam, Tanzim Ahad, Sajedul Talukder (2025). "A Comprehensive Survey of Federated Learning for Edge AI: Recent Trends and Future Directions." Preprint.
Download Paper

AI-in-the-Loop: Privacy Preserving Real-Time Scam Detection and Conversational Scam-baiting by Leveraging LLMs and Federated Learning

Published in PETS 2026, 2026

Scams exploiting real-time social engineering—such as phishing, impersonation, and phone fraud—remain a persistent and evolving threat across digital platforms. Existing defenses are largely reactive, offering limited protection during active interactions.

Recommended citation: Ismail Hossain, Sai Puppala, Md Jahangir Alam, Sajedul Talukder (2026). "AI-in-the-Loop: Privacy Preserving Real-Time Scam Detection and Conversational Scam-baiting by Leveraging LLMs and Federated Learning." PETS 2026.
Download Paper

LAMDA: A Longitudinal Android Malware Benchmark for Concept Drift Analysis

Published in ICLR 2026, 2026

Machine learning (ML)-based malware detection systems often fail to account for the dynamic nature of real-world training and test data distributions.

Recommended citation: Md Ahsanul Haque, Ismail Hossain, Md Mahmuduzzaman Kamol, Md Jahangir Alam, Suresh Kumar Amalapuram, Sajedul Talukder, Mohammad Saidur Rahman (2026). "LAMDA: A Longitudinal Android Malware Benchmark for Concept Drift Analysis." ICLR 2026.
Download Paper

The Art of the Jailbreak: Formulating Jailbreak Attacks for LLM Security Beyond Binary Scoring

Published in Preprint (arXiv), 2026

Jailbreak attacks – adversarial prompts that bypass LLM alignment through purely linguistic manipulation – pose a growing operational security threat, yet the field lacks large-scale, reproducible infrastructure for generating, categorizing, and evaluating them systematically.

Recommended citation: Ismail Hossain, Tanzim Ahad, Md Jahangir Alam, Sai Puppala, Syed Bahauddin Alam, Sajedul Talukder (2026). "The Art of the Jailbreak: Formulating Jailbreak Attacks for LLM Security Beyond Binary Scoring." Preprint (arXiv).
Download Paper

SkillVetBench: LLM-as-Judge for Multi-Dimensional Security Risk Evaluation in Open-Source LLM Agent Skills

Published in Preprint (arXiv), 2026

Open-source LLM agent ecosystems are growing rapidly, yet the security of community-contributed skills - modular tool definitions that extend agent capabilities - remains largely unvetted.

Recommended citation: Ismail Hossain, Sai Puppala, Md Jahangir Alam, Tanzim Ahad, Sajedul Talukder (2026). "SkillVetBench: LLM-as-Judge for Multi-Dimensional Security Risk Evaluation in Open-Source LLM Agent Skills." Preprint (arXiv).
Download Paper

Agent-Fence: Mapping Security Vulnerabilities Across Deep Research Agents

Published in AAAI 2026 Summer Symposium, 2026

Large language models are becoming deep agents that plan, persist state, and invoke tools, shifting safety failures from unsafe text to unsafe trajectories.

Recommended citation: Sai Puppala, Ismail Hossain, Md Jahangir Alam, Yoonpyo Lee, Jay Yoo, Tanzim Ahad, Syed Bahauddin Alam, Sajedul Talukder (2026). "Agent-Fence: Mapping Security Vulnerabilities Across Deep Research Agents." AAAI 2026 Summer Symposium.
Download Paper

HALLPERM: Exposing the Safety Illusion in LLM Tool Use via Implicit Privilege Escalation and Semantic Risk

Published in AAAI 2026 Summer Symposium, 2026

Large language model (LLM) agents increasingly rely on external tools via structured schemas, yet the safety implications of under-specified tool interfaces remain poorly understood.

Recommended citation: Md Jahangir Alam, Tanzim Ahad, Ismail Hossain, Sai Puppala, Yoonpyo Lee, Syed Bahauddin Alam, Sajedul Talukder (2026). "HALLPERM: Exposing the Safety Illusion in LLM Tool Use via Implicit Privilege Escalation and Semantic Risk." AAAI 2026 Summer Symposium.
Download Paper

Semantic Intent Fragmentation: A Single-Shot Compositional Attack on Multi-Agent AI Pipelines

Published in AAAI 2026 Summer Symposium, 2026

We introduce Semantic Intent Fragmentation (SIF), a new attack class against large language model (LLM) orchestration systems.

Recommended citation: Tanzim Ahad, Ismail Hossain, Md Jahangir Alam, Sai Puppala, Yoonpyo Lee, Syed Bahauddin Alam, Sajedul Talukder (2026). "Semantic Intent Fragmentation: A Single-Shot Compositional Attack on Multi-Agent AI Pipelines." AAAI 2026 Summer Symposium.
Download Paper

When Safety Geometry Collapses: Fine-Tuning Vulnerabilities in Agentic Guard Models

Published in AAAI 2026 Summer Symposium, 2026

A guard model fine-tuned on entirely benign data can lose all safety alignment– not through adversarial manipulation, but through standard domain specialization.

Recommended citation: Ismail Hossain, Sai Puppala, Jannatul Ferdaus, Md Jahangir Alam, Tanzim Ahad, Yoonpyo Lee, Syed Bahauddin Alam, Sajedul Talukder (2026). "When Safety Geometry Collapses: Fine-Tuning Vulnerabilities in Agentic Guard Models." AAAI 2026 Summer Symposium.
Download Paper

Influence Is Not Authority: When Causal Guardrail Signals Make Legitimate Tool Use Look Like an Attack in Tool-Using LLM Agents

Published in Preprint (arXiv), 2026

The key limitation of current state-of-the-art influence-based guardrails is that they do not reliably distinguish a legitimate, user-authorized action from a malicious, unauthorized action when both rely on external tool information.

Recommended citation: Tanzim Ahad, Ismail Hossain, Md Jahangir Alam, Sai Puppala, Syed Bahauddin Alam, Sajedul Talukder (2026). "Influence Is Not Authority: When Causal Guardrail Signals Make Legitimate Tool Use Look Like an Attack in Tool-Using LLM Agents." Preprint (arXiv).
Download Paper

Guard-to-Action: Multilingual Long-Context Safety in Tool-Using Agents

Published in ICLR 2027 (under review), 2026

A multilingual guard-to-agent evaluation across 98 languages, five runtime guards, and three tool-using agents, measuring whether guardrail bypasses propagate into restricted-tool selection at 8K, 20K, and 32K context.

Recommended citation: Md Jahangir Alam, Ismail Hossain, Tanzim Ahad, Sai Puppala, Sajedul Talukder (2026). "Guard-to-Action: Multilingual Long-Context Safety in Tool-Using Agents." Under review at ICLR 2027.

Lost in Speech: Measuring the Multilingual Cross-Modal Safety Gap in Speech-Language Models

Published in ARR submitted (NAACL), 2026

Unsafe compliance stays near zero across six speech-language models while comprehension collapses outside English and outside text — so up to 76% of nominally safe spoken responses are safe only because the model did not understand the request.

Recommended citation: Md Jahangir Alam, Tanzim Ahad, Ismail Hossain, Sai Puppala, Sajedul Talukder (2026). "Lost in Speech: Measuring the Multilingual Cross-Modal Safety Gap in Speech-Language Models." Submitted to ACL Rolling Review (NAACL).

PolyAEGIS: Benchmarking Large Language Model Safety Across 100 Languages

Published in ICLR 2027 (under review), 2026

A massively multilingual LLM safety benchmark: 10,362 Gold-quality prompts across 100 languages, six generation models, and a three-judge consensus protocol over 61,536 scored responses.

Recommended citation: Md Jahangir Alam, Tanzim Ahad, Ismail Hossain, Sai Puppala, Syed Bahauddin Alam, Sajedul Talukder (2026). "PolyAEGIS: Benchmarking Large Language Model Safety Across 100 Languages." Under review at ICLR 2027.

SafeCache: Security-Aware Context Eviction for Tool-Using Agents

Published in MLSys (under review), 2026

Context eviction is not behaviorally neutral: dropping a trusted authorization policy raises unauthorized tool use from 0% to 66.7–82.5%, and a deterministic non-evictable tag eliminates the failure.

Recommended citation: Md Jahangir Alam, Tanzim Ahad, Ismail Hossain, S M Abrar Jahin, Sai Puppala, Sajedul Talukder (2026). "SafeCache: Security-Aware Context Eviction for Tool-Using Agents." Under review at MLSys.

When Guardrails Fail: Multilingual Long-Context Safety in Tool-Using Agents

Published in NeurIPS 2026 Workshop (LCFM), 2026

A multilingual guard-to-agent evaluation across 98 languages and 186K+ instances, measuring whether guardrail bypasses propagate into restricted-tool selection under 8K and 32K contexts.

Recommended citation: Md Jahangir Alam, Ismail Hossain, Tanzim Ahad, Sai Puppala, Sajedul Talukder (2026). "When Guardrails Fail: Multilingual Long-Context Safety in Tool-Using Agents." NeurIPS 2026, Long-Context Foundation Models (LCFM).
Download Paper

talks

teaching

Machine Learning / NLP Labs and Linux Programming

Graduate Teaching Assistant, Southern Illinois University Carbondale, School of Computing, 2024

Spring 2024. Led machine learning and NLP labs along with Linux programming sessions; mentored and graded graduate and undergraduate students.

Automata Theory

Graduate Teaching Assistant, University of Texas at El Paso, Department of Computer Science, 2025

Spring 2025. Mentored and graded graduate and undergraduate students in the Automata course.

Computer Networks

Graduate Teaching Assistant, University of Texas at El Paso, Department of Computer Science, 2025

Fall 2025 and Spring 2026. Mentoring and grading graduate and undergraduate students in the Computer Networks course.