Agentic AI Security: Threats, Defenses, Evaluation, and Open Challenges
Безопасность агентного искусственного интеллекта: угрозы, методы защиты, оценка и открытые проблемы
2026-01-01
SCID: 54.1/g7mzam3h
Discuss with AI
agentic AI securityautonomous agentslarge language modelssecurity threatstool use
Figures from the paper
Abstract (AI)
Agentic AI systems powered by large language models (LLMs) and endowed with planning, tool use, memory, and autonomy, are emerging as powerful, flexible platforms for automation. Their ability to autonomously execute tasks across web, software, and physical environments creates new and amplified security risks, distinct from both traditional AI safety and conventional software security. This survey outlines a taxonomy of threats specific to agentic AI, reviews recent benchmarks and evaluation methodologies, and discusses defense strategies from both technical and governance perspectives.We synthesize current research and highlight open challenges, aiming to support the development of secure-by-design agent systems.
Key Findings
1
Agentic AI systems introduce security risks amplified by planning, tool use, memory, and autonomy across web, software, and physical environments.
2
Existing defenses span technical mechanisms and governance strategies, supporting a secure-by-design approach to agent development.
3
Major open challenges remain in securely deploying and evaluating autonomous agent systems.
4
The survey develops a taxonomy of agentic AI-specific threats and reviews corresponding benchmarks and evaluation methodologies.
5
These risks are distinct from both traditional AI safety concerns and conventional software security threats.
Research Object
Agentic AI systems powered by large language models, with planning, tool use, memory, and autonomy
Research Subject
Security threats, defenses, evaluation methods, and open challenges associated with autonomous agentic AI operation across web, software, and physical environments
Publication Details
Publication Date
2026-01-01
Journal
Publisher
ISSN
Cited by
128
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai11
Intriguing properties of neural networks2013
A Comprehensive Overview of Large Language Models2025
Generative Agents: Interactive Simulacra of Human Behavior2023
GPT-4 Technical Report2023
Evaluating Large Language Models Trained on Code2021
Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection2023
Empowering biomedical discovery with AI agents2024
InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents2024
Current state of LLM Risks and AI Guardrails2024
TRiSM for Agentic AI: A review of Trust, Risk, and Security Management in LLM-based Agentic Multi-Agent Systems2026
Evaluation and Benchmarking of LLM Agents: A Survey2025