EUVIMEDEUVIMED entdecken
Crossref

SECURITY THREATS TO AI AGENTS: REVIEW ON THREATS, DEFENSE STRATEGIES AND THREAT MODELLING

Geethu Merin Mathew, Kapil Shukla, Krishna Modi

Journal of Digital Security and Forensics · 06.06.2026 · 3 (1) · 56-66

DOI: 10.29121/digisecforensics.v3.i1.2026.98 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Digital Security and Forensics, online erschienen am 06.06.2026. Verfasst von Geethu Merin Mathew und 2 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Artificial intelligence agents are no longer simply language-based systems; rather, they are becoming autonomous entities that can reason, use tools, interact with external environments, and collaborate with each other. This has led to a significant increase in the scope of security threats associated with artificial intelligence agents, shifting the focus from machine learning-based security threats to more complex system-level security threats. This review aims to highlight the emerging security threats associated with artificial intelligence agents by bringing together various recent research works related to artificial intelligence agents, language model security, and autonomous decision-making systems. This study highlights the various significant types of security threats associated with artificial intelligence agents and analyze these types of threats in detail. These types of security threats include input-level attacks, model-level attacks, tool-level attacks, memory and state manipulation attacks,environment-level attacks, and multi-agent level attacks. The review aims to highlight various types of defense mechanisms available against these types of security threats, including input sanitizing, tool filtering, isolation architectures, safety agents, and policy-based tool execution. The review aims to highlight various recent advancements in threat modeling techniques like ASTRIDE, ATFAA, MAESTRO, and risk scoring techniques related to critical security threats like prompt injection attacks, memory poisoning attacks, unauthorized tool usage, and data exfiltration. The analysis has shown that currently available defense mechanisms against various types of security threats associated with artificial intelligence agents are limited and insufficient in nature; therefore, a systems security approach is required to address artificial intelligence agent security issues by considering architectural safeguards, standard evaluation benchmarks, better management of memories and tool usage in artificial intelligence agents, and better alignment of security requirements with governance requirements.

Zitieren

APA

Mathew, G. M., Shukla, K., & Modi, K. (2026). SECURITY THREATS TO AI AGENTS: REVIEW ON THREATS, DEFENSE STRATEGIES AND THREAT MODELLING. Journal of Digital Security and Forensics, 3(1), 56-66. https://doi.org/10.29121/digisecforensics.v3.i1.2026.98

Vancouver

Mathew GM, Shukla K, Modi K. SECURITY THREATS TO AI AGENTS: REVIEW ON THREATS, DEFENSE STRATEGIES AND THREAT MODELLING. Journal of Digital Security and Forensics. 2026;3(1):56-66. doi:10.29121/digisecforensics.v3.i1.2026.98

BibTeX
@article{Mathew2026,
  author = {Mathew, Geethu Merin and Shukla, Kapil and Modi, Krishna},
  title = {SECURITY THREATS TO AI AGENTS: REVIEW ON THREATS, DEFENSE STRATEGIES AND THREAT MODELLING},
  journal = {Journal of Digital Security and Forensics},
  year = {2026},
  volume = {3},
  number = {1},
  pages = {56--66},
  doi = {10.29121/digisecforensics.v3.i1.2026.98}
}
Kostenlos starten →