CoChat

Large Model-Based Agents: State-of-the-Art, Cooperation Paradigms, Security and Privacy, and Future Trends

Open access
Wang, Yuntao, Pan, Yanghe, Su, Zhou, Deng, Yi, Zhao, Quan, Du, Linkang, Luan, Tom H., Kang, Jiawen, Niyato, Dusit · 2026 · IEEE Communications Surveys & Tutorials
Cited by 68

The DOI matches its Crossref record; 66 of the 66 references it cites were found in a registry or an index.

Abstract

With the rapid advancement of large models (LMs), the development of general-purpose intelligent agents powered by LMs has become a reality. It is foreseeable that in the near future, LM-driven general AI agents will serve as essential tools in production tasks, capable of autonomous communication and collaboration without human intervention. This paper investigates scenarios involving the autonomous collaboration of future LM agents. We review the current state of LM agents, the key technologies enabling LM agent collaboration, and the security and privacy challenges they face during cooperative operations. To this end, we first explore the foundational principles of LM agents, including their general architecture, key components, enabling technologies, and modern applications. We then discuss practical collaboration paradigms from data, computation, and knowledge perspectives to achieve connected intelligence among LM agents. After that, we analyze the security vulnerabilities and privacy risks associated with LM agents, particularly in multi-agent settings, examining underlying mechanisms and reviewing current and potential countermeasures. Lastly, we propose future research directions for building robust and secure LM agent ecosystems.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1109/comst.2025.3576176

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 164

  1. Shavit, 2023, Practices for governing agentic AI systems
  2. 10.1023/A:1015008417172 · DOI
  3. Paglieri, BALROG: Benchmarking agentic LLM and VLM reasoning on games, 1
  4. Cheng, 2024, Exploring large language model based intelligent agents: Definitions, methods, and prospects, arXiv:2401.03428 · arXiv
  5. 10.57020/ject.1297961 · DOI
  6. Wu, AutoGen: Enabling next-gen LLM applications via multi-agent conversation, 1
  7. 10.18653/v1/2024.findings-emnlp.270 · DOI
  8. Ahn, Do as I can, not as I say: Grounding language in robotic affordances, 1
  9. Nakano, 2022, WebGPT: Browser-assisted question-answering with human feedback, arXiv:2112.09332 · arXiv
  10. 10.52202/079017-0088 · DOI
  11. 10.1109/IROS58592.2024.10802229 · DOI
  12. Deng, PentestGPT: Evaluating and harnessing large language models for automated penetration testing, 847
  13. 10.1109/MWC.016.2300600 · DOI · Paper page
  14. Marro, 2025, Agora protocol: Scalable and reliable communication between your agents
  15. Zhang, VPGTrans: Transfer visual prompt generator across LLMs, 36, 20299
  16. Ma, LLM-pruner: On the structural pruning of large language models, 36, 21702
  17. 10.1609/aaai.v38i17.29860 · DOI
  18. 10.1109/MWC.005.2400019 · DOI
  19. Fedus, 2022, Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity, J. Mach. Learn. Res., 23, 1
  20. Zou, PoisonedRAG: Knowledge poisoning attacks to retrieval-augmented generation of large language models, 1
  21. Carlini, Quantifying Memorization across neural language models, 1
  22. Liu, 2023, Prompt injection attack against LLM-integrated applications, arXiv:2306.05499 · arXiv
  23. Fang, Local model poisoning attacks to Byzantine-robust federated learning, 1605
  24. Rando, Universal jailbreak backdoors from poisoned human feedback, 1
  25. Carlini, Extracting training data from large language models, 2633
  26. 10.1109/COMST.2024.3353265 · DOI · Paper page
  27. 10.24963/ijcai.2024/890 · doi:10.24963/ijcai.2024/890 · Paper page
  28. Durante, 2024, Agent AI: Surveying the horizons of multimodal interaction, arXiv:2401.03568 · arXiv
  29. 10.1109/comst.2025.3527641 · DOI · Paper page
  30. 10.1109/OJCOMS.2024.3456549 · DOI · Paper page
  31. Mei, 2024, AIOS: LLM agent operating system, arXiv:2403.16971 · arXiv
  32. Wei, Chain-of-thought prompting elicits reasoning in large language models, 35, 24824
  33. Yao, Tree of thoughts: Deliberate problem solving with large language models, 1
  34. Yao, ReAct: Synergizing reasoning and acting in language models, 1
  35. Shinn, Reflexion: Language agents with verbal reinforcement learning, 8634
  36. Wang, Voyager: An open-ended embodied agent with large language models, 1
  37. Hu, AVIS: Autonomous visual information seeking with large language model agent, 867
  38. Lewis, 2020, Retrieval-augmented generation for knowledge-intensive NLP tasks, Proc. NeurIPS, 33, 9459
  39. Kang, Knowledge-augmented reasoning distillation for small language models in knowledge-intensive tasks, 36, 1
  40. Liu, 2024, From LLM to conversational agent: A memory enhanced architecture with fine-tuning of large language models, arXiv:2401.02777 · arXiv
  41. Huang, Language models as zero-shot planners: Extracting actionable knowledge for embodied agents, 9118
  42. Driess, PaLM-E: An embodied multimodal language model, 202, 8469
  43. Talebirad, 2023, Multi-agent collaboration: Harnessing the power of intelligent LLM agents, arXiv:2306.03314 · arXiv
  44. Hong, MetaGPT: Meta programming for a multi-agent collaborative framework, 1
  45. 2025, Unitree G1
  46. 2025, Figure 02 humanoid robot
  47. 2025, Baidu Apollo launches Apollo ADFM large model
  48. 10.18653/v1/2023.emnlp-main.13 · DOI · Paper page
  49. Jinxin, 2023, CGMI: Configurable general multi-agent interaction framework, arXiv:2308.12503 · arXiv
  50. 10.1145/3637528.3671620 · DOI
  51. 10.1109/MNET.2024.3376419 · DOI
  52. Xu, 2024, AutoAttacker: A large language model guided system to implement automatic cyber-attacks, arXiv:2403.01038 · arXiv
  53. Guo, 2025, DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning, arXiv:2501.12948 · arXiv
  54. 2025, AgenticFlow AI
  55. 2025, Dify.AI
  56. 2025, RAGFlow
  57. 10.1109/MNET.2024.3435752 · DOI · Paper page
  58. Hao, ToolkenGPT: Augmenting frozen language models with massive tools via tool embeddings, 1
  59. 10.18653/v1/2023.findings-emnlp.691 · DOI · Paper page
  60. 10.1109/IROS58592.2024.10801334 · DOI
  61. Wan, Knowledge fusion of large language models, 1
  62. 10.1109/TIV.2024.3396450 · DOI
  63. 10.1609/aaai.v37i13.26928 · doi:10.1609/aaai.v37i13.26928
  64. 10.1109/TMC.2011.64 · doi:10.1109/tmc.2011.64
  65. 10.1109/mnet.2024.3442880 · DOI
  66. 10.1109/ICCV51070.2023.02137 · DOI
  67. 10.1109/comst.2024.3516819 · DOI
  68. 10.1145/1293731.1293735 · DOI · Paper page
  69. 10.1145/191246.191322 · DOI · Paper page
  70. 2024, Model context protocol (MCP)
  71. 2024, Agent network protocol (ANP)
  72. 10.1186/s43058-022-00380-5 · DOI
  73. 10.1109/tmc.2025.3546466 · DOI
  74. 10.1109/ICDCS60910.2024.00013 · DOI
  75. 10.18653/v1/2024.acl-long.363 · DOI
  76. Chan, ChatEval: Towards better LLM-based evaluators through multi-agent debate, 1
  77. Xu, 2024, Cached model-as-a-resource: Provisioning large language model agents for edge intelligence in space-air-ground integrated networks, arXiv:2403.05826 · arXiv
  78. Liu, A dynamic LLM-powered agent network for task-oriented agent collaboration, 1
  79. 10.1109/OJCS.2023.3300321 · DOI · Paper page
  80. Brown, Language models are few-shot learners, 33, 1877
  81. 10.1007/978-3-030-70604-3_4 · DOI
  82. Shen, Mixture-of-experts meets instruction tuning: A winning combination for large language models, 1
  83. 10.1525/9780520940420-020 · DOI · Paper page
  84. 10.1109/TWC.2022.3221778 · DOI
  85. 10.24963/ijcai.2024/3 · DOI
  86. 10.1109/TCNS.2023.3242359 · doi:10.1109/tcns.2023.3242359
  87. 10.1002/9781119863663.ch8 · DOI
  88. 10.1109/ICCV48922.2021.00148 · DOI
  89. 10.1609/aaai.v37i4.25555 · DOI · Paper page
  90. Qian, Scaling large language model-based multi-agent collaboration, 1
  91. Zhong, Seeking neural nuggets: Knowledge transfer in large language models from a parametric perspective, 1
  92. 10.18653/v1/2022.findings-naacl.26 · DOI
  93. 10.18653/v1/2022.findings-acl.220 · DOI
  94. 10.18653/v1/2023.acl-long.792 · DOI · Paper page
  95. Shen, Joint language semantic and structure embedding for knowledge graph completion, 1965
  96. 10.18653/v1/2024.findings-acl.400 · DOI
  97. 10.18653/v1/2020.emnlp-main.437 · DOI · Paper page
  98. 10.1109/TKDE.2023.3325484 · DOI
  99. Xu, An LLM can fool itself: A prompt-based adversarial attack, 1
  100. Du, Stable diffusion is unstable, 1
  101. 10.1109/SP54263.2024.00102 · DOI
  102. Luo, An image is worth 1000 lies: Transferability of adversarial images across prompts on vision-language models, 1
  103. 10.1109/SP54263.2024.00123 · DOI
  104. Deng, Jailbreaker: Automated jailbreak across multiple large language model chatbots, 1
  105. Yu, Don’t listen to me: Understanding and exploring jailbreak prompts of large language models, 1
  106. Zhang, BadRobot: Jailbreaking embodied LLMs in the physical world, 1
  107. Toyer, Tensor trust: Interpretable prompt injection attacks from an online game, 1
  108. Schuster, You autocomplete me: Poisoning vulnerabilities in neural code completion, 1559
  109. Zhang, Agent security bench (ASB): Formalizing and benchmarking attacks and defenses in LLM-based agents, 1
  110. 10.18653/v1/2025.emnlp-main.1771 · DOI
  111. Chan, 2024, AgentMonitor: A plug-and-play framework for predictive and secure multi-agent systems, arXiv:2408.14972 · arXiv
  112. 10.1109/ICCV51070.2023.00423 · DOI
  113. 10.18653/v1/2024.naacl-long.171 · DOI
  114. Shi, Large language models can be easily distracted by irrelevant context, 202, 31210
  115. Liu, Training socially aligned language models on simulated social interactions, 1
  116. Liang, Adversarial example does good: Preventing painting imitation from diffusion models via adversarial examples, 202, 20763
  117. 10.18653/v1/2023.acl-industry.56 · DOI
  118. 10.18653/v1/2020.acl-main.529 · DOI
  119. Kumar, 2023, Certifying LLM safety against adversarial prompting, arXiv:2309.02705 · arXiv
  120. Phute, LLM self defense: By self examination, LLMs know they are being tricked, 1
  121. Zeng, AutoDefense: Multi-agent LLM defense against jailbreak attacks, 1
  122. 10.14722/ndss.2024.24115 · DOI
  123. Jones, Automatically auditing large language models via discrete optimization, 15307
  124. Xu, 2024, RedAgent: Red teaming large language models with context-aware autonomous language agent, arXiv:2407.16667 · arXiv
  125. Zhou, WebArena: A realistic Web environment for building autonomous agents, 1
  126. 10.18653/v1/2024.findings-naacl.217 · DOI
  127. 10.18653/v1/2021.findings-emnlp.369 · DOI
  128. 10.14722/ndss.2024.23238 · DOI
  129. 10.18653/v1/2022.findings-emnlp.148 · DOI · Paper page
  130. 10.18653/v1/2023.acl-long.709 · DOI
  131. Carlini, Extracting training data from diffusion models, 5253
  132. 10.18653/v1/2022.emnlp-main.570 · DOI · Paper page
  133. Kong, An efficient membership inference attack for the diffusion model by proximal Initialization, 1
  134. 10.52202/079017-4290 · DOI
  135. Wang, Property existence inference against generative models, 1
  136. Krishna, Thieves on sesame street! model extraction of BERT-based APIs, 1
  137. 10.1145/3576915.3616652 · DOI
  138. 10.1145/3597503.3639091 · DOI
  139. 10.18653/v1/2023.emnlp-main.189 · DOI
  140. Shen, Prompt stealing attacks against text-to-image generation models, 1
  141. Sha, 2024, Prompt stealing attacks against large language models, arXiv:2402.12959 · arXiv
  142. Mauran, 2024, Samsung bans ChatGPT, AI Chatbots after data leak blunder
  143. 10.1109/JPROC.2023.3306773 · DOI · Paper page
  144. 10.1016/j.neucom.2021.04.105 · DOI
  145. 10.1109/SP.2019.00031 · DOI · Paper page
  146. Kandpal, User inference attacks on large language models, 1
  147. 10.18653/v1/2022.emnlp-main.119 · DOI · Paper page
  148. Kandpal, Deduplicating training data mitigates privacy risks in language models, 10697
  149. 10.18653/v1/2021.findings-emnlp.102 · DOI
  150. Kim, ProPILE: Probing privacy leakage in large language models, 1
  151. 10.1109/SP.2018.00038 · DOI · Paper page
  152. 10.1145/2810103.2813677 · DOI · Paper page
  153. Kirchenbauer, A watermark for large language models, 17061
  154. 10.1109/TSC.2024.3382958 · DOI
  155. Liu, Mitigating hallucination in large multi-modal models via robust instruction tuning, 1
  156. Penedo, The RefinedWeb dataset for falcon LLM: Outperforming curated corpora with Web data only, 1
  157. Lee, Factuality enhanced language models for open-ended text generation, 1
  158. Zhang, How language model hallucinations can snowball, 1
  159. Jones, Teaching language models to hallucinate less with synthetic tasks, 1
  160. Mündler, Self-contradictory hallucinations of large language models: Evaluation, detection and mitigation, 1
  161. Chen, INSIDE: LLMs’ internal states retain the power of hallucination detection, 1
  162. Zhou, Analyzing and mitigating object hallucination in large vision-language models, 1
  163. 10.1109/CVPR52729.2023.00373 · DOI
  164. 10.1109/ICCV48922.2021.00466 · DOI

Cited by

See all 68

Metadata

Authors
Wang, Y., Pan, Y., Su, Z., Deng, Y., Zhao, Q., Du, L., Luan, T. H., Kang, J., & Niyato, D.
Venue
IEEE Communications Surveys & Tutorials
Published
2026
Volume
28
Pages
1906-1949
ISSN
  • 1553-877X (ISSN-L)
  • 2373-745X
OpenAlex
W4410985905
Cited by
68
References
66
FWCI
27.88
Type
journal article
Language
en
Open access
green
Funders
NSFC (62302387, U22A2029, U24A20237, U23A20276, 62402379)

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.