Pritom Bhowmik

AI Security Researcher/LLM Agents & Memory Security

Research Area: LLM Agent SecurityMemory PoisoningRetrieval-Time AttacksDefense Cost EvaluationMulti-Agent TrustMachine Learning

Portrait of Pritom Bhowmik
01

Research

I work on the security of memory in LLM agents: how memory gets poisoned, how attacks enter at retrieval time, and what defenses cost in accuracy and compute. More broadly, I study trustworthy multi-agent systems and the integrity of agent-to-agent communication.

02

Publications

  1. 2026
    The Price of Safety: Benign-Case Utility and Token Overhead of Memory-Poisoning Defenses in LLM Agents
    Pritom Bhowmik
    arXiv:2609.22818 [cs.CR], 2026. Under review, ARTMAN Workshop at ACSAC 2026.
    [PDF][arXiv]
  2. 2026
    Outcome-Bound Trust: Falsifiable Claims as a Security Primitive for Memory in Agent-to-Agent AI Systems
    Pritom Bhowmik
    Manuscript in preparation, 2026.
  3. 2022
    Machine Learning in Production: From Experimented ML Model to System
    Pritom Bhowmik
    ScienceOpen Preprints, 2022.
    [PDF][DOI]
  4. 2021
    A Data-Centric Approach to Improve Machine Learning Model’s Performance in Production
    Pritom Bhowmik
    International Journal of Engineering and Advanced Technology (IJEAT), 2021.
    [Paper][DOI]
  5. 2021
    Vulnerability of Neural Networks to Adversarial Attack and Defenses in Computer Vision
    Pritom Bhowmik
    IEEE Computer Society, Bangladesh Chapter, 2021. Abstract.
  6. 2019
    Research Study on Basic Understanding of Artificial Neural Networks
    Pritom Bhowmik
    Global Journal of Computer Science and Technology: D Neural & Artificial Intelligence, 19(4), pp. 5–7, 2019.
    [PDF][Paper]
03

Research & Teaching Experience

Research

  • Researcher, AI SecuritySep 2025 – Present
    Memory Security in LLM Agent Systems
  • Graduate Researcher, Data Science LabJan 2024 – May 2025
    Montclair State University · Advisor: Dr. Hao Liu

Teaching

  • Adjunct Instructor, Computer ScienceSep 2025 – Jul 2026
    Monroe University, New Rochelle, NY
    Undergraduate and graduate courses in computer science, data science, and database systems.
04

Current Research

2026 · Under review, ARTMAN @ ACSAC 2026

The Price of Safety

  • Measures what memory-poisoning defenses cost when no attack is present, across 4 LLM backbones and 2 blinded judges (κ ≥ 0.84).
  • Write-time defenses: no measurable cost.
  • Read-time reranker: −4.4 points accuracy; quarantines legitimate memories on 33.6% of items.

[PDF][arXiv]

2026 · Manuscript in preparation

Outcome-Bound Trust

  • Uses falsifiable claims as a security primitive for memory shared between agents.
  • Asks what a receiving agent should trust when another agent’s output becomes its memory.
05

Education

  • M.S., Data ScienceMay 2025
    Montclair State University

    GPA 3.97 / 4.00
    Alpha Epsilon Lambda Honor Society

  • B.Tech., Computer ScienceJul 2020
    Institute of Engineering and Management, Kolkata