A Comparative Survey of Security Risks in AI Systems: From LLMs to AI Agents and Embodied Agents
Сравнительный обзор рисков безопасности в системах искусственного интеллекта: от больших языковых моделей до ИИ-агентов и воплощённых агентов
2026-08-01
SCID: 54.1/8wmdcqkk
Discuss with AI
AI agentsAI security risksembodied agentslarge language modelssecurity assessment frameworks
Figures from the paper
Abstract (AI)
Rapid AI development across industries raises pressing security and privacy risks. This work presents a unified comparison of large language models, AI agents, and embodied agents, introducing a taxonomy of risks spanning data, models, systems, content, and applications, alongside a catalog of 24 specific threats. We contrast attack surfaces and methods across the three system types to reveal common patterns and distinctive vulnerabilities. We also survey mainstream AI security assessment frameworks and evaluate how relevant laws and regulations currently address these risks. Finally, we outline concrete directions for future research and practice aimed at building robust, secure agent ecosystems.
Key Findings
1
Comparing attack surfaces and methods reveals both shared patterns and vulnerabilities distinctive to each system category.
2
It introduces a five-layer risk taxonomy covering data, models, systems, content, and applications.
3
The survey provides a unified security comparison of large language models, AI agents, and embodied agents.
4
The survey reviews AI security assessment frameworks and evaluates how existing laws and regulations address identified risks, while proposing future research directions.
5
The work catalogs 24 specific security and privacy threats across the three AI system types.
Research Object
large language models, AI agents, and embodied agents
Research Subject
comparative security and privacy risks, attack surfaces, vulnerabilities, assessment frameworks, and regulatory coverage across the three AI system types
Publication Details
Publication Date
2026-08-01
Journal
Publisher
ISSN
Cited by
2
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai13
A survey on large language model based autonomous agents2024
🧜Siren’s Song in the AI Ocean: A Survey on Hallucination in Large Language Models2025
The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery2024
A survey on large language model (LLM) security and privacy: The Good, The Bad, and The Ugly2024
Prompt Injection attack against LLM-integrated Applications2023
Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection2023
A Survey of Methods for Explaining Black Box Models2018
Regulating ChatGPT and other Large Generative AI Models2023
PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models2024
AI Agents Under Threat: A Survey of Key Security Challenges and Future Pathways2025
InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents2024
Risks of AI scientists: prioritizing safeguarding over autonomy2025
Zero-shot Persuasive Chatbots with LLM-Generated Strategies and Information Retrieval2024