The UMBC Cyber Defense Lab presents

Self-Defending Ledgers: Automating Distributed Ledger Security Using Multi-Agent Reinforcement Learning and Game Theory

Md Tariqul Islam 'Pavel'
Assistant Professor of Cybersecurity
UMBC Department of Information Systems

12–1pm EDT, Friday, October 3, 2025 via Webex

Distributed ledger technologies (DLTs) continue to face significant security challenges. While attackers constantly adapt their strategies, governance mechanisms often remain static. Our work addresses this critical gap by introducing a framework for self-defending ledgers, where nodes enforce ledger security through adaptive governance driven by multi-agent reinforcement learning (MARL) grounded in game-theoretic principles. We model DLT consensus as a repeated Bayesian game, in which participants hold probabilistic beliefs about peer behavior, allowing agents to make strategic decisions under partial observability of adversarial actions. Our framework enables nodes to model, detect, and respond to a wide range of malicious behaviors, including bribery, selfish mining, equivocation, Sybil attacks, and collusive voting, by continuously updating Bayesian trust beliefs and governance policies based on network observations. We formally prove that networks with an honest majority reach stable equilibria and provide bounds on adversarial influence. Experiments across five major protocols show that agents effectively identify attacks with high accuracy while substantially reducing adversarial success. This work demonstrates the potential of game-theoretic MARL to provide robust, self-adaptive security in varied DLT environments, paving the way for resilient and autonomous ledger governance.

Md Tariqul Islam 'Pavel' is an assistant professor of cybersecurity in UMBC's Department of Information Systems. His research centers on the security, efficiency, and fault-tolerance of distributed computing systems, with a strong emphasis on blockchain, cloud, and vehicular networks. He develops formal models, algorithms, and protocols that address critical vulnerabilities in decentralized ecosystems, spanning inter-blockchain communication, smart contract migration, and trustworthy governance. His work combines cryptography, game theory, and system design to build scalable, resilient infrastructures. He earned his PhD and MS from the University of Kentucky and BS from the University of Dhaka, Bangladesh.

Recording of the event

UMBC's Center for Social Science Scholarship will hold a workshop series this Fall on generative AI, LLMs, and computational social science methods.  This series will begin with the basics of computing in R and how to use generative AI/LLMs in social science research workflows.  Each session will focus on helping faculty and students become comfortable with deploying AI and LLM models in their research, while also gaining a deeper understanding of the ethical, equity, and environmental implications of these models.  

This series is supported through the Elkins Professorship, features several speakers, and will continue into the spring 2026 semester. The four sessions this Fall are:

• Introduction to R, R Projects, R Markdown, and the Basics of LLMs, Dr. Eric Stokan, 12-1:30 pm, October 3, in PUP 451,
• Inductive Analysis of Texts with Embeddings, Dr. Dustin Stoltz (Lehigh University), 12-1:30 pm, November 5, Commons 329
• Dealing with unstructured data (text, images), APIs, and API calls to LLMs, Dr. Eric Stokan, 12-1:30pm, November 7, PUP 451
• Prompt-engineering and Fine-tuning in R, Dr. Eric Stokan, 12-1:30pm, December 12, PUP 451

Register for the sessions here.

Cosponsored by DoIT, the Center for Scalable Data and Computational Science, and CGC-SCIPE.