Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap
Open accessPublished version of a preprint
The DOI matches its Crossref record; 146 of the 146 references it cites were found in a registry or an index; 43 of the 159 sources its claims cite were checked for support.
Abstract
This survey reviews agentic AI and LLM-driven approaches for autonomous IoT cybersecurity and identifies trustworthy multi-agent collaboration and resource-hardened edge agents as high-priority directions.
Overview
CoChat Summary
TL;DR
This PRISMA-style systematic survey synthesizes 153 peer-reviewed studies (2020–2026) on LLM- and agentic-AI-driven IoT cybersecurity, organizing them along a four-pillar taxonomy of agent architecture, reasoning strategy, action scope, and deployment topology. It concludes that the field is dominated by detection-only, single-dataset demonstrations, and proposes a 2026 roadmap centered on federated agentic learning, verifiable reasoning, trustworthy multi-agent collaboration, and resource-hardened edge agents.
Key findings
- Of the 41 studies coded as primary IoT-IDS work, 38 (~93%) operate in detection-only mode; the corpus contains no system conforming to MCP/ANP/A2A agent-interoperability protocols, and only 5 of 153 studies (~3%) report any compound multi-step failure mode (error propagation, context poisoning, tool-use amplification).
- Cross-dataset transfer degrades accuracy by 10–20 percentage points across at least nine primary studies, and only 2 of 13 agentic vulnerability-discovery studies report a verified false-discovery rate.
- Reported frontiers (Table 2): edge/on-device SLMs at 70–400 ms p50 with ~99% on Edge-IIoTset; fog/gateway 7–13B models at 300–900 ms with ~98% on CICIoT2023; cloud 70B+ models at 1–5 s end-to-end with ~98% prediction accuracy. Prompt-injection susceptibility is largely unmeasured at the edge against a CyberSecEval baseline of 26–41%.
- The companion kit's deterministic stub scored F1 = 0.83 on synthetic Edge-IIoTset-style traffic but only 0.06 accuracy on real Edge-IIoTset flows — reported as a negative control quantifying the synthetic-to-real gap.
Methods
- PRISMA 2020 protocol over IEEE Xplore, ACM Digital Library, and MDPI journals, January 2020–May 2026 (final search 29 May 2026); 487 candidate records → 63 duplicates removed → 424 title/abstract screened → 218 excluded as out of scope → 206 full-text screened → 53 excluded → 153 included studies.
- Corpus distribution: 58 IEEE, 39 ACM DL, 56 MDPI; by year, 9 (2020), 14 (2021), 18 (2022), 24 (2023), 36 (2024), 38 (2025), 14 (2026 to cut-off).
- Each study coded along seven dimensions (identifier, IoT layer/topology, agent architecture, reasoning strategy, cybersecurity focus, benchmarks, limitations); first-author screening with senior-author re-screening of a 15% random sample (Cohen's κ = 0.84, 90.4% agreement; seven adjusted inclusion decisions).
Limitations
- Stated: (1) corpus restricted to three venues, omitting USENIX Security, IEEE S&P, ACM CCS, NDSS and Elsevier/Springer journals; (2) screening primarily by one reviewer with only a 15% spot-check rather than two independent reviewers; (3) paywalled papers extracted from abstract/metadata only; (4) temporal upper bound of May 2026; (5) English-only corpus; (6) the companion reproducibility kit is v0.3.0 scaffolding, never validated in production, with only a deterministic stub run on synthetic traffic.
- Evident from the text: 56 of 153 included studies (37%) come from MDPI, the publisher of the manuscript itself — a disclosed but unresolved venue-concentration bias; the paper also notes that the top-tier-venue narrative scan was selected by expert judgement rather than reproducible screening and does not contribute to the 153-paper count.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
27 Valid · 18 Not valid · 61 Not checked · 154 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“SOAR and threathunting surveys [27,28] focus on workflow automation rather than autonomous reasoning.”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 4. Related Surveys and Positioning
What the sources say
- Disputedcites [27] Srinivas, S., Kirk, B., Zendejas, J., and Alzahrani, N. (2025). AI-Augmented SOC: A Survey of LLMs and Agents for Security Automation. J. Cybersecur. Priv., 5. sourceExperimental — accuracy not yet measured
The Introduction (page unavailable) describes this paper as a survey of LLM and AI-agent applications across SOC functions, not specifically a SOAR or threat-hunting survey. Its Capability Maturity Model section explicitly examines progression from manual to fully autonomous operations, including independent complex analysis. That contradicts the characterization that it focuses on workflow automation rather than autonomous reasoning. (LLM, conf=0.87)
Checked by a reasoning model — may be wrong.
- Disputedcites [28] Kurnia, R., Brata, Z.A., Nelistiani, G.A., Heo, S., and Kim, H. (2025). Toward Robust Security Orchestration and Automated Response in SOC with a Hyper-Automation Approach Using Agentic AI. Information, 16. sourceExperimental — accuracy not yet measured
Our copy of 10.3390/info16050365 is not a SOAR/threat-hunting survey but a framework proposal (Abstract; Key Contributions) whose central argument runs counter to the claim: Section 2.2 states that traditional no-code/low-code SOAR is limited by static playbooks whereas 'contemporary SOAR solutions leverage LLMs to dynamically generate code... these AI-powered systems integrate contextual awareness and reasoning to respond in real time', and Sections 2.3/2.5.2 develop agentic AI and LLM 'multi-step reasoning' for autonomous incident response. The paper does discuss workflow/playbook automation (Section 2.2.4; Introduction on playbook challenges; References 6-12 on threat-hunting automation and playbook generation), which touches the 'workflow automation' half, but it explicitly foregrounds autonomous reasoning rather than omitting it. Thus the characterization 'focus on workflow automation rather than autonomous reasoning' is contradicted by our copy. (LLM, conf=0.72)
Checked by a reasoning model — may be wrong.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The 5G/6G transition compounds the problem by introducing network slicing-logical partitions whose isolation guarantees become themselves the target of slice-hopping attacks”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 2.1. The IoT Security Landscape
What the sources say
- Supportedcites [38] Allaw, Z., Zein, O., and Ahmad, A.-M. (2025). Cross-Layer Security for 5G/6G Network Slices: An SDN, NFV, and AI-Based Hybrid Framework. Sensors, 25. source
The Introduction describes network slicing as a 5G/6G feature that creates isolated virtual networks but raises concerns about cross-slice attacks. Section 2.2 adds that poor slice isolation can permit unauthorized access and cross-contamination between services, supporting the substance of the cited sentence. (LLM, conf=0.88)
- Partly supportedcites [39] Dias, J., Pinto, P., Santos, R., and Malta, S. (2025). 5G Network Slicing: Security Challenges, Attack Vectors, and Mitigation Approaches. Sensors, 25. sourceExperimental — accuracy not yet measured
The Introduction (page unavailable) describes network slicing as independent virtual networks on shared infrastructure, and the Discussion (page unavailable) reports cross-slice access and lateral movement enabled by poor isolation. These passages support the substance of attacks crossing slice boundaries, but the paper addresses 5G slicing; it does not establish the claimed 5G/6G transition as the cause of compounded IoT security risks or identify those attacks by the term “slice-hopping.” (LLM, conf=0.82)
- Partly supportedcites [48] Botez, R., Zinca, D., and Dobrota, V. (2025). Redefining 6G Network Slicing: AI-Driven Solutions for Future Use Cases. Electronics, 14. sourceExperimental — accuracy not yet measured
Our copy squarely supports the definitional half of the sentence: the Introduction (Sec. 1) states that network slicing is 'already fundamental to 5G but envisioned to play an even more critical role in 6G' and creates 'virtualized, application-specific network segments on shared infrastructure' for the proliferating IoT, and Sec. 1.1 defines a slice as 'an independent, logically isolated virtual network that operates on shared physical infrastructure'. The attributed security point is absent: no passage in our copy mentions slice-hopping or any attack that targets slice isolation; on the contrary, the Introduction frames isolation as protective ('slicing also bolsters security by isolating sensitive data flows ... mitigating risks of data breaches or unauthorized access'). Searches for 'slice hopping attack', 'cross-slice', 'adversary moving between slices', and 'IoT security challenges/threats' returned only slicing-performance and requirement material (Sec. 1.1, Sec. 2 Related Works, Tables 3-5). So the source supports the transition-and-partition framing but not the claim that isolation guarantees become an attack target. (LLM, conf=0.70)
- Partly supportedcites [49] Karahan, S.N., Güllü, M., Karhan, D., Çimen, S., Osmanca, M.S., and Barışçı, N. (2025). Realistic Performance Assessment of ML Algorithms for 6G Network Slicing: A Dual-Methodology Approach with Explainable AI. Electronics, 14. sourceExperimental — accuracy not yet measured
Our copy supports the framing half of the claim: the Introduction (p.1) and §2.1 describe network slicing as 'virtualized end-to-end logical network slices on shared physical infrastructure' established in 5G and evolving into the 6G paradigm, and §2 Related Work and Background lists 'ensuring slice isolation... protecting data security and privacy' among the key challenges of the 5G/6G vertical scenarios, with §2.1 (summarising Cunha et al. [23]) naming 'slice isolation, inter-slice handover' as significant security challenges. However, the specific element attributed here — that these isolation guarantees become the target of 'slice-hopping attacks' — is nowhere stated: searches for 'slice-hopping attack', 'hopping between slices', and 'attacker bypass isolation' returned only semantic neighbours (slice classification performance, handover algorithms, SHAP tables), never an attack class exploiting inter-slice isolation. The paper is also a slice-classification performance study, so the isolation-threat material is background reported from other works rather than a finding of this paper. (LLM, conf=0.68)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“And they raise governance questions about accountability when autonomous decisions cause harm [27,28].”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 2.3. The Agentic AI Paradigm
What the sources say
- Supportedcites [27] Srinivas, S., Kirk, B., Zendejas, J., and Alzahrani, N. (2025). AI-Augmented SOC: A Survey of LLMs and Agents for Security Automation. J. Cybersecur. Priv., 5. source
Section 7 (Threats to Validity; page unavailable) identifies accountability for misaligned decisions and the ethical implications of semi-autonomous escalation as poorly explored. Section 4 (Capability Maturity Model; page unavailable) says greater autonomy shifts human roles toward governance and requires stronger ethical standards. Together, these support the claim as a paraphrase of a governance concern, not a claim that harm has occurred. (LLM, conf=0.87)
- Partly supportedcites [28] Kurnia, R., Brata, Z.A., Nelistiani, G.A., Heo, S., and Kim, H. (2025). Toward Robust Security Orchestration and Automated Response in SOC with a Hyper-Automation Approach Using Agentic AI. Information, 16. sourceExperimental — accuracy not yet measured
Our copy does touch the governance/accountability point, but only in the literature-review passage of Section 2.5.2 (LLMs and Agentic AI in Cybersecurity Domain), where it says the ethical implications of deploying autonomous AI systems in cybersecurity are explored by Radanliev et al., 'emphasizing the need for transparency and accountability' — i.e. the paper reports someone else's treatment rather than establishing the point itself. Nothing in our copy ties accountability questions to autonomous decisions that cause harm: the nearest risk discussion (Section 5, Discussion) is confined to misclassification/false positives, training-data quality, interpretability and the need for human intervention to validate critical steps, and Section 4.5 describes the agent deferring to human intervention rather than causing harm. The 'when autonomous decisions cause harm' qualifier in the citing sentence is therefore unsupported by this source. (LLM, conf=0.72)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Agents inherit and amplify the weaknesses of the underlying LLM-hallucination, prompt sensitivity, and contextual drift-and propagate them across longer decision chains [20,21].”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 2.3. The Agentic AI Paradigm
What the sources say
- Partly supportedcites [20] Jaffal, N.O., Alkhanafseh, M., and Mohaisen, D. (2025). Large Language Models in Cybersecurity: A Survey of Applications, Vulnerabilities, and Defense Techniques. AI, 6. sourceExperimental — accuracy not yet measured
Section 5 identifies hallucinations as an LLM risk, and §5.3 describes how prompts can manipulate LLM outputs. Section 4.5.2 also notes contextual limitations. The paper does not establish that agents amplify these weaknesses or propagate them across longer decision chains; no page numbers are available in our copy. (LLM, conf=0.82)
- Partly supportedcites [21] Atlam, H.F. (2025). LLMs in Cyber Security: Bridging Practice and Education. Big Data Cogn. Comput., 9. sourceExperimental — accuracy not yet measured
Section 6.1, “Model Reliability and Accuracy” (page not provided), reports LLM hallucinations and reliability that varies with prompt structure and context. The paper also describes difficulty with multi-step reasoning, but the passages do not establish that agents amplify these weaknesses or propagate them across longer decision chains; they do not specifically establish contextual drift in agents. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“IoV security has been reviewed by Maoudi et al. [102] and Junejo et al. [103] without an LLM agent lens.”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 4. Related Surveys and Positioning
What the sources say
- Not checked — no full text of the sourcecites [102] Das, 2026, Towards the Safety of Intelligent Transportation: A Survey on the Security Challenges and Mitigations in Internet of Vehicles (IoV), ACM Trans. Sens. Netw. source
- Partly supportedcites [103] Alalwany, E., and Mahgoub, I. (2024). Security and Trust Management in the Internet of Vehicles (IoV): Challenges and Machine Learning Solutions. Sensors, 24. sourceExperimental — accuracy not yet measured
The Introduction and Section 4 (Related Surveys; page not provided) show that this paper reviews IoV security and discusses earlier IoV-security surveys, with no LLM-agent lens evident in the cited passages. But this source is by Alalwany and Mahgoub, and the passages do not establish that the named Maoudi and Junejo papers are the reviews described in the claim. (LLM, conf=0.82)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
66 more claims · 8 with their cited sources checked
Create a free account to see all 260 claims and to verify the 44 references still open.
- Probably not a substantial claim. 105
- It may state two separate claims as one. 131
- The quote only partly states this claim. 2
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 5.4.3. Cloud-Centric
“Cloud topology removes resource constraints but introduces unbounded latency (hundreds of milliseconds end-to-end)”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > Model
“Cloud-centric 70B+ params, frontier API models 1-5 s end-to-end”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > Sanity-check validation run.
“No number from Appendix A is used anywhere in the body of this survey”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > Sanity-check validation run.
“the stub contains no language model, and its figures (binary attack-vs.-benignF 1 = 0.83) validate harness functionality only”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 5.3.2. Response Orchestration
“CyberAlly [27] is reported to have cut MTTR from 8 h to 90 min”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > Agentic AI does not eliminate the
“| CyberAlly [24] | Multi-step ReAct with analyst hand-off | SOC ticketing + EDR API | Vendor-reported MTTR delta |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 4. Related Surveys and Positioning
“Yao et al. [20] survey LLM applications across eight cybersecurity domains”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > References
“20. Jaffal, N.O.; Alkhanafseh, M.; Mohaisen, D. Large Language Models in Cybersecurity: A Survey of Applications, Vulnerabilities, and Defense Techniques.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“the 5% prompt-injection ceiling is a tenth of the 26-41% CyberSecEval baseline for production LLMs”
§ Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap > 9.4. Resource-Hardened Edge Agents
Reported 0.1 · recomputed 0.192308
2 more findings
Create a free account to see every finding of this check.
Citations
Cites
Communication-Efficient Learning of Deep Networks from Decentralized Data
McMahan, H. B., Moore, E., Ramage, D., Hampson, S., & Arcas, B. A. Y. (2016) · arXiv · doi:10.48550/arxiv.1602.05629 · Cited by 14,215
Chain-Of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q. V., & Zhou, D. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1800 · Cited by 4,505
CICIoT2023: A real-time dataset and benchmark for large-scale attacks in IoT environment
Neto, E. C. P., Dadkhah, S., Ferreira, R., Zohourian, A., Lu, R., & Ghorbani, A. A. (2023) · doi:10.20944/preprints202305.0443.v1 · Cited by 1,235
CICIoT2023: A Real-Time Dataset and Benchmark for Large-Scale Attacks in IoT Environment
Neto, E. C. P., Dadkhah, S., Ferreira, R., Zohourian, A., Lu, R., & Ghorbani, A. A. (2023) · Sensors · doi:10.3390/s23135941 · Cited by 1,194
Machine Learning DDoS Detection for Consumer Internet of Things Devices
Doshi, R., Apthorpe, N., & Feamster, N. (2018) · 2018 IEEE Security and Privacy Workshops (SPW) · doi:10.1109/spw.2018.00013 · Cited by 680
Greshake, K., Abdelnabi, S., Mishra, S., Endres, C., Holz, T., & Fritz, M. (2023) · Proceedings of the 16th ACM Workshop on Artificial Intelligence and Security · doi:10.1145/3605764.3623985 · Cited by 667 · cited 3×
“The earlier direct-prompt-injection concept was named by Perez and Ribeiro in a NeurIPS 2022 ML Safety Workshop Best Paper, but as that work has not appeared in a formal peer-reviewed proceedings, we cite the peer-reviewed Greshake follow-up [94] as the foundational reference.”
InfluentialWhen LLMs meet cybersecurity: a systematic literature review
Zhang, J., Bu, H., Wen, H., Liu, Y., Fei, H., Xi, R., Li, L., Yang, Y., Zhu, H., & Meng, D. (2025) · Cybersecurity · doi:10.1186/s42400-025-00361-w · Cited by 383 · cited 2×
“The closest excluded-venue comparators—the Springer and Elsevier systematic reviews discussed in Section 3.8 [99,100]—are likewise partial with respect to this intersection: both treat LLMs-for-security broadly, but neither centers agentic control loops, IoT deployment topology, or a companion…”
InfluentialBotnet Detection in the Internet of Things using Deep Learning Approaches
McDermott, C. D., Majdani, F., & Petrovski, A. V. (2018) · 2018 International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2018.8489489 · Cited by 350
Zhou, X., Liang, W., Li, W., Yan, K., Shimizu, S., & Wang, K. I.-K. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3130434 · Cited by 325 · cited 2×
“The first is the conventional model from non-LLM intrusion detection: adversaries craft traffic perturbations that evade ML classifiers [142–144].”
Artificial intelligence risk management framework :
National Institute of Standards and Technology (US) (2024) · NIST Trustworthy and Responsible AI · doi:10.6028/nist.ai.600-1 · Cited by 274
Large Language Model guided Protocol Fuzzing
Meng, R., Mirchev, M., Böhme, M., & Roychoudhury, A. (2024) · Proceedings 2024 Network and Distributed System Security Symposium · doi:10.14722/ndss.2024.24556 · Cited by 246 · cited 1×
“On vulnerability discovery, KARONTE (Redini et al., IEEE S&P 2020 [107]) and ChatAFL (Meng et al., NDSS 2024 [108]) represent the systems-security firmware-and protocol-fuzzing tradition we flagged as under-represented: KARONTE is a non-LLM multi-binary taint-tracking baseline, whereas ChatAFL uses…”
Nwakanma, C. I., Ahakonye, L. A. C., Njoku, J. N., Odirichukwu, J. C., Okolie, S. A., Uzondu, C., Ndubuisi Nweke, C. C., & Kim, D.-S. (2023) · Applied Sciences · doi:10.3390/app13031252 · Cited by 202
Zhang, C., Costa-Perez, X., & Patras, P. (2022) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2021.3137084 · Cited by 195
Hermosilla, P., Berríos, S., & Allende, H. (2025) · Applied Sciences · doi:10.3390/app15137329 · Cited by 171
PentestAgent: Incorporating LLM Agents to Automated Penetration Testing
Shen, X., Wang, L., Li, Z., Chen, Y., Zhao, W. J., Sun, D., Wang, J., & Ruan, W. (2025) · ACM Asia Conference on Computer and Communications Security (ASIA CCS) · doi:10.1145/3708821.3733882 · Cited by 170 · cited 2×
“This is the topology used by most threat-hunting [24,25], penetration testing [32,33,35], and CTI extraction [29–31 , 130”
Karonte: Detecting Insecure Multi-binary Interactions in Embedded Firmware
Redini, N., Machiry, A., Wang, R., Spensky, C., Continella, A., Shoshitaishvili, Y., Kruegel, C., & Vigna, G. (2020) · IEEE Symposium on Security and Privacy (S&P) · doi:10.1109/sp40000.2020.00036 · Cited by 168 · cited 1×
“On vulnerability discovery, KARONTE (Redini et al., IEEE S&P 2020 [107]) and ChatAFL (Meng et al., NDSS 2024 [108]) represent the systems-security firmware-and protocol-fuzzing tradition we flagged as under-represented: KARONTE is a non-LLM multi-binary taint-tracking baseline, whereas ChatAFL uses…”
Reconstructed Graph Neural Network With Knowledge Distillation for Lightweight Anomaly Detection
Zhou, X., Wu, J., Liang, W., Wang, K. I.-K., Yan, Z., Yang, L. T., & Jin, Q. (2024) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3389714 · Cited by 145 · cited 2×
“Small language models (SLMs) of 0.5–3 billion parameters dominate this tier, typically delivered via quantization (4-bit or 8-bit), pruning, and distillation [72,73,79–84].”
InfluentialGetting pwn’d by AI: Penetration Testing with Large Language Models
Happe, A., & Cito, J. (2023) · Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering · doi:10.1145/3611643.3613083 · Cited by 144
Towards Human-AI Teaming to Mitigate Alert Fatigue in Security Operations Centres
Chhetri, M. B., Tariq, S., Singh, R., Jalalvand, F., Paris, C. L., & Nepal, . (2024) · ACM Transactions on Internet Technology · doi:10.1145/3670009 · Cited by 139
A Lightweight IoT Intrusion Detection Model Based on Improved BERT-of-Theseus
Wang, Z., Li, J., Yang, S., Luo, X., Li, D., & Mahmoodi, S. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.122045 · Cited by 131 · cited 3×
“Small language models (SLMs) of 0.5–3 billion parameters dominate this tier, typically delivered via quantization (4-bit or 8-bit), pruning, and distillation [72,73,79–84].”
InfluentialAlalwany, E., & Mahgoub, I. (2024) · Sensors · doi:10.3390/s24020368 · Cited by 130 · cited 1×
“IoV security has been reviewed by Maoudi et al. [102] and Junejo et al. [103] without an LLM agent lens.”
ChatGPT for Cybersecurity: Practical Applications, Challenges, and Future Directions
Al-Hawawreh, M., Aljuhani, A., & Jararweh, Y. (2023) · Cluster Computing · doi:10.1007/s10586-023-04124-5 · Cited by 124 · cited 3×
“Worked illustrations of the genuinely overlapping cases: HuntGPT [18] = (single-agent, CoT, detection, cloud); Cur-riculumPT [36] = (multi-agent, plan-and-solve, vulnerability discovery, cloud); the SEED edge IDS [73] = (single-agent, none, detection, edge); and the hybrid LLM-IDS designs of…”
A Review of Smart Grid Anomaly Detection Approaches Pertaining to Artificial Intelligence
Burgos, M. F. G., Morato, J., & Vizcaíno-Imacaña, P. (2024) · Applied Sciences · doi:10.3390/app14031194 · Cited by 118 · cited 1×
“Defensive practice has historically relied on signature-based intrusion detection systems (IDS), supplemented by anomaly detectors based on autoencoders, isolation forests, and recurrent networks [2,50–52].”
Design and Implementation of IoT DDoS Attacks Detection System based on Machine Learning
Chen, Y.-W., Sheu, J., Kuo, Y.-C., & Cuong, N. V. (2020) · European Conference on Networks and Communications · doi:10.1109/eucnc48522.2020.9200909 · Cited by 111 · cited 1×
“…exhibits distinct attack surfaces: at the perception layer, firmware exploits and command injection [40,41]; at the network layer, traffic flooding, protocol fuzzing, and lateral movement [42–45]; at the application layer, credential abuse, supply chain compromise, and data exfiltration [46,47].”
Demystifying RCE Vulnerabilities in LLM-Integrated Apps
Liu, T., Deng, Z., Meng, G., Li, Y., & Chen, K. (2024) · ACM SIGSAC Conference on Computer and Communications Security (CCS) · doi:10.1145/3658644.3690338 · Cited by 109 · cited 1×
“…measurement our Section 8.2 argues the IoT corpus lacks, while “Demystifying RCE Vulnerabilities in LLM-Integrated Apps” (Liu et al., ACM CCS 2024 [111]) shows that prompt injection can escalate to remote code execution in tool-augmented applications—the concrete realization of the tool-hijacking…”
InfluentialIntelligent Detection of IoT Botnets Using Machine Learning and Deep Learning
Kim, J., Shim, M., Hong, S., Shin, Y., & Choi, E. (2020) · Applied Sciences · doi:10.3390/app10197009 · Cited by 106
Internet of Things Botnet Detection Approaches: Analysis and Recommendations for Future Research
Wazzan, M., Algazzawi, D., Bamasaq, O., Albeshri, A., & Cheng, L. (2021) · Applied Sciences · doi:10.3390/app11125713 · Cited by 102
E-GRACL: an IoT intrusion detection system based on graph neural networks
Lin, L., Zhong, Q., Qiu, J., & Liang, Z. (2024) · The Journal of Supercomputing · doi:10.1007/s11227-024-06471-5 · Cited by 95
Wang, Y., Yu, Z., Wu, J., Wang, C., Zhou, Q., & Hu, J. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3387328 · Cited by 94
GNN-IDS: Graph Neural Network based Intrusion Detection System
Sun, Z., Teixeira, A. M. H., & Toor, S. (2024) · Proceedings of the 19th International Conference on Availability, Reliability and Security · doi:10.1145/3664476.3664515 · Cited by 80 · cited 1×
“Reviews on deep reinforcement learning for IoT intrusion detection [105] and graph-based IoT IDS [62,63] address complementary AI paradigms without LLM integration.”
Blockchain-Enabled Zero Trust Architecture for Privacy-Preserving Cybersecurity in IoT Environments
Aleisa, M. A. (2025) · IEEE Access · doi:10.1109/access.2025.3529309 · Cited by 77 · cited 1×
“…exhibits distinct attack surfaces: at the perception layer, firmware exploits and command injection [40,41]; at the network layer, traffic flooding, protocol fuzzing, and lateral movement [42–45]; at the application layer, credential abuse, supply chain compromise, and data exfiltration [46,47].”
Alabbadi, A., & Bajaber, F. (2025) · Sensors · doi:10.3390/s25030847 · Cited by 77
Evaluation and Benchmarking of LLM Agents: A Survey
Mohammadi, M., Li, Y., Lo, J., & Yip, W. (2025) · Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 · doi:10.1145/3711896.3736570 · Cited by 74
A Zero-Trust Architecture for Remote Access in Industrial IoT Infrastructures
Federici, F., Martintoni, D., & Senni, V. (2023) · Electronics · doi:10.3390/electronics12030566 · Cited by 74
Benchmarking Large Language Models for Log Analysis, Security, and Interpretation
Karlsen, E., Luo, X., Zincir-Heywood, N., & Heywood, M. (2024) · Journal of Network and Systems Management · doi:10.1007/s10922-024-09831-x · Cited by 71 · cited 9×
“CyberSecEval (versions 1–3) and CyberSOCEval, summarized in [19,86], benchmark LLM safety properties: insecure coding suggestions, cyberattack helpfulness, prompt-injection resistance, code-interpreter abuse, and autonomous offensive capability.”
InfluentialJaffal, N. O., Alkhanafseh, M., & Mohaisen, D. (2025) · AI · doi:10.3390/ai6090216 · Cited by 69
Fatema, K., Dey, S. K., Anannya, M., Khan, R. T., Rashid, M., Su, C., & Mazumder, R. (2025) · Future Internet · doi:10.3390/fi17060234 · Cited by 69
The Emerged Security and Privacy of LLM Agent: A Survey with Case Studies
He, F., Zhu, T., Ye, D., Liu, B., Zhou, W., & Yu, P. S. (2024) · arXiv · doi:10.48550/arxiv.2407.19354 · Cited by 60
Wisanwanichthan, T., & Thammawichai, M. (2025) · Computers · doi:10.3390/computers14070291 · Cited by 59 · cited 1×
“%) operate in detection-only mode, with representative examples spanning hybrid LLM-IDS architectures, edge transformers, and BERT-based federated detectors [16,17,65,71–74, 85”
Anomaly traffic detection in IoT security using graph neural networks
Gao, M., Wu, L., Li, Q., & Chen, W. (2023) · Journal of Information Security and Applications · doi:10.1016/j.jisa.2023.103532 · Cited by 58
The Emerged Security and Privacy of LLM Agent: A Survey with Case Studies
He, F., Zhu, T., Ye, D., Liu, B., Zhou, W., & Yu, P. S. (2026) · ACM Computing Surveys · doi:10.1145/3773080 · Cited by 58
ZTA-IoT: A Novel Architecture for Zero-Trust in IoT Systems and an Ensuing Usage Control Model
Ameer, S., Praharaj, L., Sandhu, R. S., Bhatt, S., & Gupta, M. (2024) · ACM Transactions on Privacy and Security · doi:10.1145/3671147 · Cited by 57 · cited 1×
“…exhibits distinct attack surfaces: at the perception layer, firmware exploits and command injection [40,41]; at the network layer, traffic flooding, protocol fuzzing, and lateral movement [42–45]; at the application layer, credential abuse, supply chain compromise, and data exfiltration [46,47].”
(2023) · doi:10.52202/075280-2687 · Cited by 57
Deep Reinforcement Learning for Intrusion Detection in IoT: A Survey
Gueriani, A., Kheddar, H., & Mazari, A. C. (2023) · 2023 2nd International Conference on Electronics, Energy and Measurement (IC2EM) · doi:10.1109/ic2em59347.2023.10419560 · Cited by 56 · cited 1×
“Reviews on deep reinforcement learning for IoT intrusion detection [105] and graph-based IoT IDS [62,63] address complementary AI paradigms without LLM integration.”
Intrusion Detection in IoT Networks Using Dynamic Graph Modeling and Graph-Based Neural Networks
Villegas-Ch, W. E., Govea, J., Navarro, A. M., & Játiva, P. P. (2025) · IEEE Access · doi:10.1109/access.2025.3559325 · Cited by 56 · cited 1×
“Increasingly, researchers have turned to graph neural networks (GNNs) that exploit the topological structure of IoT traffic [59–64] and to federated learning that allows distributed devices to train shared anomaly detectors without exposing raw data [57,65,66].”
An Adaptive Intrusion Detection System in the Internet of Medical Things Using Fuzzy-Based Learning
Alalhareth, M., & Hong, S.-C. (2023) · Sensors · doi:10.3390/s23229247 · Cited by 52 · cited 3×
“At the perception/network boundary, line-rate non-LLM detectors trained on CICIoMT2024 [114,131–133] run on network taps (detection scope; non-agentic baseline; edge topology) and flag suspect flows.”
Zhu, S., Xu, X., Zhao, J., & Xiao, F. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3310794 · Cited by 51
Fatema, K., Dey, S. K., Anannya, M., Khan, R. T., Rashid, M., Chunhua, S., & Mazumder, R. (2025) · doi:10.20944/preprints202503.1902.v1 · Cited by 51
GNN-Based Network Traffic Analysis for the Detection of Sequential Attacks in IoT
Altaf, T., Wang, X., Ni, W., Yu, G., Liu, R. P., & Braun, R. (2024) · Electronics · doi:10.3390/electronics13122274 · Cited by 50 · cited 1×
“Reviews on deep reinforcement learning for IoT intrusion detection [105] and graph-based IoT IDS [62,63] address complementary AI paradigms without LLM integration.”
Ismail, Kurnia, R., Brata, Z. A., Nelistiani, G. A., Heo, S., Kim, H., & Kim, H. (2025) · Information · doi:10.3390/info16050365 · Cited by 50
IoT Anomaly Detection to Strengthen Cybersecurity in the Critical Infrastructure of Smart Cities
Villegas-Ch, W., Govea, J., & Jaramillo-Alcazar, A. (2023) · Applied Sciences · doi:10.3390/app131910977 · Cited by 48 · cited 1×
“…when transferred cross-dataset, and we identified at least nine primary studies in the corpus that report this degradation when their model is evaluated on a second IoT dataset (the specific studies are enumerated in the bracketed citation list immediately following) [2,3,14,15,51,64,65,77,78].”
Georgiades, M., & Hussain, F. (2025) · Electronics · doi:10.3390/electronics14163218 · Cited by 47
KnowCTI: Knowledge-based cyber threat intelligence entity and relation extraction
Wang, G., Liu, P., Huang, J., Bin, H., Wang, X., & Zhu, H. (2024) · Computers & Security · doi:10.1016/j.cose.2024.103824 · Cited by 47 · cited 1×
“CyberNER-LLM [29], KnowCTI [30], and segment-level CTI NER frameworks [31] extract entities and relations from unstructured threat reports with accuracies up to 98% on cybercrime-forum corpora.”
Smart Contract Vulnerability Detection: The Role of Large Language Model (LLM)
Boi, B., Esposito, C., & Lee, S. (2024) · ACM SIGAPP Applied Computing Review · doi:10.1145/3687251.3687253 · Cited by 41 · cited 1×
“Multi-agent smart-contract auditors [119,121–123 , Compared to coverage-guided fuzzers such as AFL++ and libFuzzer—which excel at exploring branch space within a single binary but treat every input byte as opaque—LLM-driven discovery contributes three capabilities.”
Orman, A. (2025) · Applied Sciences · doi:10.3390/app15063121 · Cited by 41
Xia, Y., Dittler, D., Jazdi, N., Chen, H., & Weyrich, M. (2024) · 2024 IEEE 29th International Conference on Emerging Technologies and Factory Automation (ETFA) · doi:10.1109/etfa61755.2024.10710900 · Cited by 41 · cited 1×
“The interplay of digital twins and cyber deception [128] suggests an emerging architecture in which a digital twin of the production IoT environment serves both as a high-fidelity decoy and as a simulation testbed for response playbooks [138,139].”
AI-Augmented SOC: A Survey of LLMs and Agents for Security Automation
Srinivas, S., Kirk, B., Zendejas, J., Espino, M., Boskovich, M., Bari, A., Dajani, K., & Alzahrani, N. (2025) · Journal of Cybersecurity and Privacy · doi:10.3390/jcp5040095 · Cited by 39 · cited 10×
“SOAR-LLM [28] dynamically generates and adapts response playbooks rather than following static templates; CyberAlly [27] is reported to have cut MTTR from 8 h to 90 min (vendor case-study figures, not an independent benchmark) in simulated SOC environments.”
InfluentialHeydari, V., & Nyarko, K. (2025) · Electronics · doi:10.3390/electronics14163249 · Cited by 38
A Systematic Review of Data-Driven Attack Detection Trends in IoT
Haque, S., El-Moussa, F., Komninos, N., & Muttukrishnan, R. (2023) · Sensors · doi:10.3390/s23167191 · Cited by 38
A Novel Lightweight IoT Intrusion Detection Model Based on Self-Knowledge Distillation
Wang, Z., Zhou, R., Yang, S., He, D., & Chan, S. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3533092 · Cited by 38 · cited 1×
“Speculative decoding, quantization, and KV-cache-reuse optimizations partially close the gap [79,80,84] but cannot eliminate it.”
Loumachi, F. Y., Ghanem, M. C., & Ferrag, M. A. (2025) · Computers · doi:10.3390/computers14020067 · Cited by 37
Ngo, T., Yin, J., Ge, Y.-F., & Wang, H. (2025) · Information · doi:10.3390/info16060499 · Cited by 37
Liu, P., Zheng, Y., Sun, C. P., Qin, C., Fang, D., Liu, M., & Sun, L. (2023) · International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) · doi:10.1145/3623278.3624759 · Cited by 36 · cited 2×
“Third, taint-aware prioritization: the agent can ingest the output of a static analyzer (e.g., LATTE or FITS [41]) and steer the fuzzer toward sinks an attacker can reach.”
Comparing Machine Learning and Deep Learning for IoT Botnet Detection
Gandhi, R., & Li, Y. (2021) · IEEE International Conference on Smart Computing · doi:10.1109/smartcomp52413.2021.00053 · Cited by 34 · cited 1×
“BoT-IoT, N-BaIoT, and IoT-23 remain widely cited foundational datasets for botnet-specific research [10–15,140 Each dataset has notable limitations.”
A Comprehensive Survey on Generative AI Solutions in IoT Security
López Delgado, J. L., & López Ramos, J. A. (2024) · Electronics · doi:10.3390/electronics13244965 · Cited by 33
Advanced Smart Contract Vulnerability Detection via LLM-Powered Multi-Agent Systems
Wei, Z., Sun, J., Sun, Y., Liu, Y., Wu, D., Zhang, Z., Zhang, X., Li, M., Liu, Y., Li, C., et al. (2025) · IEEE Transactions on Software Engineering · doi:10.1109/tse.2025.3597319 · Cited by 31 · cited 2×
“To quantify the gap rather than merely assert it: across the full 153-paper corpus, only five studies ( ≈ 3%)—[22,71,125] for error propagation and [26,87] for tool-use amplification—report observing a compound failure mode at all, and none of the 153 reports a dedicated quantitative metric for any…”
Humans and Automation: Augmenting Security Operation Centers
Tilbury, J., & Flowerday, S. (2024) · Journal of Cybersecurity and Privacy · doi:10.3390/jcp4030020 · Cited by 30 · cited 3×
“AI-Augmented AI-Augmented SOC surveys [27,54] cover agent-based SOC automation in detail but are oriented toward enterprise networks rather than IoT edge constraints.”
Redefining 6G Network Slicing: AI-Driven Solutions for Future Use Cases
Botez, R., Zinca, D., & Dobrota, V. (2025) · Electronics · doi:10.3390/electronics14020368 · Cited by 29 · cited 1×
“The 5G/6G transition compounds the problem by introducing network slicing—logical partitions whose isolation guarantees become themselves the target of slice-hopping attacks [38,39,48,49].”
Cross-Layer Security for 5G/6G Network Slices: An SDN, NFV, and AI-Based Hybrid Framework
Allaw, Z., Zein, O., & Ahmad, A.-M. (2025) · Sensors · doi:10.3390/s25113335 · Cited by 29 · cited 2×
“The 5G/6G transition compounds the problem by introducing network slicing—logical partitions whose isolation guarantees become themselves the target of slice-hopping attacks [38,39,48,49].”
Daniel, N., Kaiser, F. K., Giladi, S., Sharabi, S., Moyal, R., Shpolyansky, S., Murillo, A., Elyashar, A., & Puzis, R. (2025) · Big Data and Cognitive Computing · doi:10.3390/bdcc9020023 · Cited by 29
BARTPredict: Empowering IoT Security with LLM-Driven Cyber Threat Prediction
Diaf, A., Korba, A. A., Elislem Karabadji, N., & Ghamri-Doudane, Y. (2024) · GLOBECOM 2024 - 2024 IEEE Global Communications Conference · doi:10.1109/globecom52923.2024.10901770 · Cited by 28 · cited 6×
“BARTPredict achieves 98% accuracy in predicting IoT attacks on the BoT-IoT benchmark (80/20 stratified train/test split as reported in the primary study) by fine-tuning a BART encoder on temporal traffic sequences [71].”
InfluentialAgentic AI Security: Threats, Defenses, Evaluation, and Open Challenges
Chhabra, A., Datta, S., Nahin, S. K., & Mohapatra, P. (2026) · IEEE Access · doi:10.1109/access.2026.3675554 · Cited by 28 · cited 6×
“The second is the development of agentic AI: software systems that wrap LLM reasoning in perception, planning, memory, tool-use, and multi-agent communication loops, enabling them to autonomously execute multi-step security tasks that previously required human analysts [22–28].”
InfluentialLLMs in Cyber Security: Bridging Practice and Education
Atlam, H. F. (2025) · Big Data and Cognitive Computing · doi:10.3390/bdcc9070184 · Cited by 27 · cited 4×
“Recent surveys treat LLMs in cybersecurity generally [20,21], generative AI in IoT broadly [1], or specific sub-domains such as SOC automation [27], CTI extraction [29–31], or autonomous penetration testing [32–36].”
InfluentialA Comprehensive Study of Mozi Botnet
Tu, T., Qin, J., Zhang, H., Chen, M., Xu, T., & Huang, Y. (2022) · International Journal of Intelligent Systems · doi:10.1002/int.22866 · Cited by 27
Azhari, R. G., Suryani, V., Pahlevi, R. R., & Wardana, A. A. (2022) · doi:10.1109/icoict55009.2022.9914830 · Cited by 26
5G Network Slicing: Security Challenges, Attack Vectors, and Mitigation Approaches
Dias, J., Pinto, P., Santos, R., & Malta, S. (2025) · Sensors · doi:10.3390/s25133940 · Cited by 26 · cited 2×
“The 5G/6G transition compounds the problem by introducing network slicing—logical partitions whose isolation guarantees become themselves the target of slice-hopping attacks [38,39,48,49].”
Wu, X., Tian, Y., Chen, Y., Ye, P., Cui, X., Jia, J., Li, S., Liu, J., & Niu, W. (2025) · Applied Sciences · doi:10.3390/app15169096 · Cited by 26 · cited 6×
“(i) Pipeline coordination: agents are arranged as a directed acyclic graph through which artifacts flow, exemplified by CurriculumPT’s pentest stages [36] and Pen-testAgent’s reconnaissance-exploit-post-exploit chain [33].”
InfluentialPenHeal: A Two-Stage LLM Framework for Automated Pentesting and Optimal Remediation
Huang, J., & Zhu, Q. (2023) · Proceedings of the Workshop on Autonomous Cybersecurity · doi:10.1145/3689933.3690831 · Cited by 25 · cited 6×
“AutoPentest [32,35] and Pen-Heal [32] use plan-and-solve to enumerate attack stages before exploit selection.”
InfluentialA Multi-Agent System for Cybersecurity Threat Detection and Correlation Using Large Language Models
Hmimou, Y., Tabaa, M., Khiat, A., & Hidila, Z. (2025) · IEEE Access · doi:10.1109/access.2025.3602681 · Cited by 25 · cited 3×
“The collaborative IoT intrusion detection framework of Liu et al. [23] reports over 90% classification accuracy across three benchmark datasets, attributable to specialization gains.”
HoneyLLM: A Large Language Model-Powered Medium-Interaction Honeypot
Fan, W., Yang, Z., Liu, Y., Qin, L., & Liu, J. (2025) · Lecture Notes in Computer Science · doi:10.1007/978-981-97-8801-9_13 · Cited by 24 · cited 3×
“LLM-powered honeypots generate context-aware, dynamic responses that sustain attacker engagement orders of magnitude longer than their predecessors [126,127].”
From Edge Transformer to IoT Decisions: Offloaded Embeddings for Lightweight Intrusion Detection
Adjewa, F., Esseghir, M., & Merghem-Boulahia, L. (2026) · Sensors · doi:10.3390/s26020356 · Cited by 23 · cited 7×
“Edge-deployed DistilBERT and TinyBERT models classify IoT intrusions at over 99% accuracy with sub-70-millisecond inference latency [72,73].”
InfluentialBalasubramanian, P., Seby, J., & Kostakos, P. (2023) · 2023 IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata59044.2023.10386976 · Cited by 22
Efficient LLMs for Edge Devices: Pruning, Quantization, and Distillation Techniques
Agrawal, R., Kumar, H., & Lnu, S. R. (2025) · 2025 International Conference on Machine Learning and Autonomous Systems (ICMLAS) · doi:10.1109/icmlas64557.2025.10968787 · Cited by 21 · cited 2×
“Small language models (SLMs) of 0.5–3 billion parameters dominate this tier, typically delivered via quantization (4-bit or 8-bit), pruning, and distillation [72,73,79–84].”
InfluentialHarnessing Digital Twin Security Simulations for systematic Cyber Threat Intelligence
Dietz, M., Schlette, D., & Pernul, G. (2022) · 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC) · doi:10.1109/compsac54236.2022.00129 · Cited by 20 · cited 1×
“The interplay of digital twins and cyber deception [128] suggests an emerging architecture in which a digital twin of the production IoT environment serves both as a high-fidelity decoy and as a simulation testbed for response playbooks [138,139].”
Kavian, A., Pourhashem Kallehbasti, M. M., Kazemi, S., Firouzi, E., & Ghafari, M. (2024) · Proceedings of the 28th International Conference on Evaluation and Assessment in Software Engineering · doi:10.1145/3661167.3661263 · Cited by 20 · cited 2×
“The cost is brittleness against unfamiliar protocols and the well-documented risk of hallucinated “vulnerabilities” that do not reproduce, both of which require human-in-the-loop verification [123,124].”
Quraishi, A., Rusho, M. A., Prasad, A., Keshta, I. M., Rivera, R., & Bhatt, M. W. (2024) · International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) · doi:10.1109/icdcece60827.2024.10548160 · Cited by 20 · cited 1×
“Defensive practice has historically relied on signature-based intrusion detection systems (IDS), supplemented by anomaly detectors based on autoencoders, isolation forests, and recurrent networks [2,50–52].”
LogRESP-Agent: A Recursive AI Framework for Context-Aware Log Anomaly Detection and TTP Analysis
Lee, J., Jeong, Y., Han, T., & Lee, T. (2025) · Applied Sciences · doi:10.3390/app15137237 · Cited by 18
Leveraging Large Language Models for Scalable and Explainable Cybersecurity Log Analysis
Palma, G., Cecchi, G., Caronna, M., & Rizzo, A. (2025) · Journal of Cybersecurity and Privacy · doi:10.3390/jcp5030055 · Cited by 17
APOLLO: A GPT-based tool to detect phishing emails and generate explanations that warn users
Desolda, G., Greco, F., & Vigano, L. (2025) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3733049 · Cited by 16 · cited 1×
“APOLLO, a GPT-4o phishing detector, generates both a classification and a human-readable explanation per email [75,76].”
Authentication and Authorization in Zero Trust IoT: A Survey
James, M., Newe, T., O’Shea, D., & O’Mahony, G. D. (2024) · Irish Signals and Systems Conference · doi:10.1109/issc61953.2024.10603175 · Cited by 16
Generating IoT Edge Network Datasets based on the TON_IoT Telemetry Dataset
Zachos, G., Essop, I., Mantas, G., Porfyrakis, K., Ribeiro, J. C., & Rodriguez, J. (2021) · 2021 IEEE 26th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks (CAMAD) · doi:10.1109/camad52502.2021.9617799 · Cited by 16 · cited 2×
“…when transferred cross-dataset, and we identified at least nine primary studies in the corpus that report this degradation when their model is evaluated on a second IoT dataset (the specific studies are enumerated in the bracketed citation list immediately following) [2,3,14,15,51,64,65,77,78].”
Nowrozy, R. (2024) · Informatics · doi:10.3390/informatics11030045 · Cited by 15
Mozi IoT Malware and Its Botnets: From Theory To Real-World Observations
Sahota, J., & Vlajic, N. (2021) · 2021 International Conference on Computational Science and Computational Intelligence (CSCI) · doi:10.1109/csci54926.2021.00181 · Cited by 14 · cited 2×
“…botnet—which by 2021 had infected over 1.5 million IoT devices, established multiple large-scale peer-to-peer botnets, and generated more attack traffic than any contemporary IoT-malware family—exemplifies how rapidly modern IoT-scale adversaries outpace conventional signature-based defense [7–9].”
InfluentialFormalizing and Benchmarking Prompt Injection Attacks and Defenses
Liu, Y., Jia, Y., Geng, R., Jia, J., & Gong, N. Z. (2023) · arXiv · doi:10.48550/arxiv.2310.12815 · Cited by 14
Enhancing loMT Security with Deep Learning Based Approach for Medical IoT Threat Detection
Kavkas, N. C., & Yildiz, K. (2025) · 2025 13th International Symposium on Digital Forensics and Security (ISDFS) · doi:10.1109/isdfs65363.2025.11012062 · Cited by 13 · cited 3×
“At the perception/network boundary, line-rate non-LLM detectors trained on CICIoMT2024 [114,131–133] run on network taps (detection scope; non-agentic baseline; edge topology) and flag suspect flows.”
To Protect the LLM Agent Against the Prompt Injection Attack with Polymorphic Prompt
Wang, Z., Nagaraja, N., Zhang, L., Bahsi, H., Patil, P., & Liu, P. (2025) · 2025 55th Annual IEEE/IFIP International Conference on Dependable Systems and Networks - Supplemental Volume (DSN-S) · doi:10.1109/dsn-s65789.2025.00037 · Cited by 13 · cited 7×
“The second model is new with LLMs: prompt injection, in which adversaries embed instructions in observations to redirect the agent itself [26,88].”
InfluentialIoT DDoS attacks detection using machine learning techniques: A Review
Ashraf, A., & Elmedany, W. M. (2021) · International Conference on Data Analytics for Business and Industry (ICDABI) · doi:10.1109/icdabi53623.2021.9655789 · Cited by 11
Jamshidi, S., Shahabi, N., Nikanjam, A., Nafi, K. W., Khomh, F., & Fung, C. (2025) · Internet of Things · doi:10.1016/j.iot.2025.101735 · Cited by 11 · cited 2×
“Jamshidi et al. [100], in Elsevier’s Internet of Things , systematically review LLM applications to IoT security and privacy through 2025 and likewise report that cross-dataset validation is rare and that production evidence lags laboratory results.”
InfluentialIoT botnet detection using deep learning
Rabhi, S., Abbes, T., & Zarai, F. (2023) · 2023 International Wireless Communications and Mobile Computing (IWCMC) · doi:10.1109/iwcmc58020.2023.10182422 · Cited by 10 · cited 2×
“Subsequent Mirai variants and other IoT-botnet families continue to reproduce this pattern at scale [10–15].”
Automated Smart Contract Vulnerability Detection using Fine-tuned Large Language Models
Yang, Z., Man, G., & Yue, S. (2023) · Proceedings of the 2023 6th International Conference on Blockchain Technology and Applications · doi:10.1145/3651655.3651658 · Cited by 10
IoT-POT: Machine Learning-based Detection of Mirai Botnet Attacks in IoT
Sharma, A., & Babbar, H. (2024) · 2024 First International Conference on Innovations in Communications, Electrical and Computer Engineering (ICICEC) · doi:10.1109/icicec62498.2024.10808530 · Cited by 10
Almalawi, A. (2025) · Sensors · doi:10.3390/s25072235 · Cited by 10
Karras, A., Giannaros, A., Amasiadi, N., & Karras, C. (2026) · Future Internet · doi:10.3390/fi18020083 · Cited by 10
Advancing IoMT Defenses: Deep Collaborative Learning for Robust Healthcare Security
Otoum, Y., Singh, P., & Nayak, A. (2024) · GLOBECOM 2024 - 2024 IEEE Global Communications Conference · doi:10.1109/globecom52923.2024.10901705 · Cited by 9
Prioritizing Vulnerability Assessment Items Using LLM Based on IoT Device Documentations
Ikegami, Y., Negishi, R., Hasegawa, K., Hidano, S., Fukushima, K., Hashimoto, K., & Togawa, N. (2024) · International Conference on Internet of Things: Systems, Management and Security (IOTSMS) · doi:10.1109/iotsms62296.2024.10710294 · Cited by 9 · cited 2×
“…[40] performs LLM-guided command-injection detection in IoT web services; LLM-powered protected-interface evasion frameworks discover broken access control in IoT firmware [120]; LLM-Boofuzz [109] orchestrates black-box protocol fuzzing with LLM-generated test cases against MQTT and CoAP stacks.”
LogRESP-Agent: A Recursive AI Framework for Context-Aware Log Anomaly Detection and TTP Analysis
Lee, J., Jeong, Y., Han, T., & Lee, T. (2025) · doi:10.20944/preprints202506.0235.v1 · Cited by 9
Interplay of Digital Twins and Cyber Deception: Unraveling Paths for Technological Advancements
Heluany, J., Amro, A., Gkioulos, V., & Katsikas, S. (2024) · Proceedings of the 2024 ACM/IEEE 4th International Workshop on Engineering and Cybersecurity of Critical Systems (EnCyCriS) and 2024 IEEE/ACM Second International Workshop on Software Vulnerability · doi:10.1145/3643662.3643955 · Cited by 8 · cited 2×
“The interplay of digital twins and cyber deception [128] suggests an emerging architecture in which a digital twin of the production IoT environment serves both as a high-fidelity decoy and as a simulation testbed for response playbooks [138,139].”
Debate-Driven Multi-Agent LLMs for Phishing Email Detection
Vy Nguyen, N. T., Childress, F. D., & Yin, Y. (2025) · 2025 13th International Symposium on Digital Forensics and Security (ISDFS) · doi:10.1109/isdfs65363.2025.11012014 · Cited by 8
Asal, B., Cakin, A., & Dilek, S. (2025) · International Congress on Human-Computer Interaction, Optimization and Robotic Applications (ICHORA) · doi:10.1109/ichora65333.2025.11017105 · Cited by 7
Agentic AI for Autonomous Cyber Threat Hunting and Adaptive Defense in Dynamic Security Environments
Sheth, A., Patel, A., Upadhyay, C., Ragothaman, H., Patil, B., & Udayakumar, S. K. (2025) · 2025 IEEE International Conference on Electro Information Technology (eIT) · doi:10.1109/eit64391.2025.11103697 · Cited by 7 · cited 5×
“SOAR-LLM [28], Agentic AI for threat hunting [24,25], and the SOC hyper-automation architecture of Alzughaibi et al. [28] follow this pattern.”
InfluentialChen, L., Deng, H., Zhang, J., Zheng, B., & Jiang, R. (2025) · Symmetry · doi:10.3390/sym17050783 · Cited by 7 · cited 4×
“CyberNER-LLM extracts entities from threat-intelligence reports with accuracies that approach human-expert performance [29,31].”
InfluentialDas, R., Zhou, L., Bi, S., Wang, T., & Hou, T. (2026) · ACM Transactions on Sensor Networks · doi:10.1145/3786592 · Cited by 7 · cited 1×
“IoV security has been reviewed by Maoudi et al. [102] and Junejo et al. [103] without an LLM agent lens.”
MT-FBERT: Malicious Traffic Detection Based on Efficient Federated Learning of BERT
Tang, J., Huang, Z., & Li, C. (2025) · Future Internet · doi:10.3390/fi17080323 · Cited by 6 · cited 4×
“Increasingly, researchers have turned to graph neural networks (GNNs) that exploit the topological structure of IoT traffic [59–64] and to federated learning that allows distributed devices to train shared anomaly detectors without exposing raw data [57,65,66].”
InfluentialDwivedi, S., Rajendran, B., Akshay, P., Acha, A., Ampatt, P., & Sudarsan, S. D. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-80020-7_27 · Cited by 6 · cited 2×
“…framework of Alzughaibi et al. [28] generates contextual playbooks per incident; IntelliSOAR enriches alerts with LLM-extracted threat-intelligence [117]; CyberAlly halved false positives from 70% to 35% (still high by SOC production standards, but a substantial relative improvement) and cut MTTR…”
Karahan, S. N., Güllü, M., Karhan, D., Çimen, S., Osmanca, M. S., & Barışçı, N. (2025) · Electronics · doi:10.3390/electronics14193841 · Cited by 6
Magani, N., & Oluoch, J. (2025) · IEEE International Conference on AI in Cybersecurity (ICAIC) · doi:10.1109/icaic63015.2025.10849039 · Cited by 5 · cited 2×
“SOAR-style response in connected and autonomous vehicles [115] and in EDR-integrated workflows [116] is now demonstrating end-to-end automation of containment actions.”
CLogLLM: A Large Language Model Enabled Approach to Cybersecurity Log Anomaly Analysis
Ren, H., Lan, K., Sun, Z., & Liao, S. (2024) · 2024 4th International Conference on Electronic Information Engineering and Computer Communication (EIECC) · doi:10.1109/eiecc64539.2024.10929078 · Cited by 5 · cited 2×
“%) operate in detection-only mode, with representative examples spanning hybrid LLM-IDS architectures, edge transformers, and BERT-based federated detectors [16,17,65,71–74, 85”
LLM-Boofuzz: Generation-Based Black-Box Fuzzing for Network Protocols via LLMs
Wang, T., Li, Y., Pan, Z., Chen, Q., Li, Z., Zhang, Y., Shen, Y., Hu, M., & Chen, Q. (2025) · Electronics · doi:10.3390/electronics14234550 · Cited by 5 · cited 3×
“…[40] performs LLM-guided command-injection detection in IoT web services; LLM-powered protected-interface evasion frameworks discover broken access control in IoT firmware [120]; LLM-Boofuzz [109] orchestrates black-box protocol fuzzing with LLM-generated test cases against MQTT and CoAP stacks.”
InfluentialAutoPen: Towards Autonomous Penetration Testing Using LLM-Powered Agents
Mei, J., Chen, S., Ma, Y., & Song, H. (2025) · International Conference on Computer Science and Application Engineering (CSAE) · doi:10.1145/3772886.3772899 · Cited by 5 · cited 2×
“This is the topology used by most threat-hunting [24,25], penetration testing [32,33,35], and CTI extraction [29–31 , 130”
Transfer learning-based Mirai botnet detection in IoT networks
Rabhi, S., Abbes, T., & Zarai, F. (2023) · 2023 International Conference on Innovations in Intelligent Systems and Applications (INISTA) · doi:10.1109/inista59065.2023.10310379 · Cited by 4
Agentic Generative AI for Automation of Cyber Security Attack Chains in Tactical MANETs
Loevenich, J. F., & Lopes, R. R. F. (2025) · 2025 IEEE 50th Conference on Local Computer Networks (LCN) · doi:10.1109/lcn65610.2025.11146384 · Cited by 4 · cited 3×
“First, dual-use research: the same ReAct loops, plan-and-solve agents, and multi-agent coordination patterns that improve defense are directly usable for unauthorized penetration testing, autonomous exploitation, and red-team automation [22,32–36 , 92].”
CyberNER-LLM: Cyber Threat Intelligence Named Entity Recognition With Large Language Model
Liu, X., Lin, W., & Ding, Z. (2026) · Lecture Notes in Computer Science · doi:10.1007/978-981-95-3543-9_28 · Cited by 4 · cited 4×
“CyberNER-LLM extracts entities from threat-intelligence reports with accuracies that approach human-expert performance [29,31].”
InfluentialRAGSec: Retrieval-Augmented Generation for Cybersecurity Threat Intelligence in Enterprise Networks
Bussari, S., Punj, P., & Balakumar, G. (2026) · IEEE International Conference on AI in Cybersecurity (ICAIC) · doi:10.1109/icaic67076.2026.11395762 · Cited by 4 · cited 1×
“This is the topology used by most threat-hunting [24,25], penetration testing [32,33,35], and CTI extraction [29–31 , 130”
Joshi, A., & Baidya, S. (2026) · Journal of Cybersecurity and Privacy · doi:10.3390/jcp6020063 · Cited by 4 · cited 2×
“The cognitive-layer survey of Ali et al. [101] explicitly addresses LLM-IoT integration but is structured around defense techniques rather than agent archi-https tectures.”
Nageshwaran, V., & Ezekiel, S. (2026) · doi:10.20944/preprints202606.0142.v1 · Cited by 4
Self-citationLeveraging Large Language Models for Autonomous Threat Detection in IoT Networks
Zhang, J. (2024) · Proceedings of the 2024 8th International Conference on Electronic Information Technology and Computer Engineering · doi:10.1145/3711129.3711223 · Cited by 3 · cited 1×
“Hybrid LLM-IDS architectures combine line-rate ML classification with LLM-driven contextual reasoning, achieving improved accuracy and substantially reduced false positives [68,74,113].”
Multi-Class Classification for IoT Attack Detection: An LLM-Based Approach
Mercan, Ö. B., Toprak, A. G., & Osmanca, M. S. (2025) · 2025 33rd Signal Processing and Communications Applications Conference (SIU) · doi:10.1109/siu66497.2025.11112284 · Cited by 3 · cited 8×
“…(single-agent, CoT, detection, cloud); Cur-riculumPT [36] = (multi-agent, plan-and-solve, vulnerability discovery, cloud); the SEED edge IDS [73] = (single-agent, none, detection, edge); and the hybrid LLM-IDS designs of [16,74,87] = (hybrid architecture, ReAct, detection-then-response, fog+cloud).”
InfluentialExploring the Potential of Anomaly Detection Through Reasoning with Large Language Models
Park, S., & Choi, D. (2025) · Applied Sciences · doi:10.3390/app151910384 · Cited by 3
Almutairi, N. (2026) · Electronics · doi:10.3390/electronics15061305 · Cited by 3 · cited 1×
“LLM-Powered Proactive Cyber-Defense [118] continuously ingests threat indicators from public platforms and ranks the resulting risks.”
A longitudinal Measurement and Analysis Study of Mozi, an Evolving P2P IoT Botnet
Wang, B., Sang, Y., Zhang, Y., Li, S., & Xu, X. (2022) · 2022 IEEE International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom) · doi:10.1109/trustcom56396.2022.00027 · Cited by 3 · cited 3×
“…botnet—which by 2021 had infected over 1.5 million IoT devices, established multiple large-scale peer-to-peer botnets, and generated more attack traffic than any contemporary IoT-malware family—exemplifies how rapidly modern IoT-scale adversaries outpace conventional signature-based defense [7–9].”
InfluentialResul, E., Țurcanu, D., & Rughinis, R. V. (2025) · RoEduNet Conference: Networking in Education and Research (RoEduNet) · doi:10.1109/roedunet68395.2025.11208322 · Cited by 2 · cited 3×
“LLM-based mapping of SIEM rules to ATT&CK techniques [129,137] (e.g., the Rule-ATT&CK Mapper framework) automates a step that previously consumed senior analyst time.”
The Impact of LLM Assistance on User Spam Detection
Morales Flores, A., Jhaveri, J., Sharma, A., Rege, S., & Munyaka, I. (2025) · Proceedings of the 2025 Workshop on Human-Centered AI Privacy and Security · doi:10.1145/3733816.3760754 · Cited by 2
Adaptive Deception Architectures: Conceptual Foundations for LLM-Powered Honeypot Systems
Érsok, M., Balogh, Á., Kail, E., & Bánáti, A. (2025) · 2025 IEEE 19th International Symposium on Applied Computational Intelligence and Informatics (SACI) · doi:10.1109/saci66288.2025.11030110 · Cited by 2 · cited 3×
“LLM-powered honeypots generate context-aware, dynamic responses that sustain attacker engagement orders of magnitude longer than their predecessors [126,127].”
How far have we been on the path of LLM-enhanced vulnerability detection
Shen, W., Yang, Y., Zhang, X., Cui, L., & Wang, Y. (2025) · Proceedings of the 2025 International Symposium on Artificial Intelligence and Computational Social Sciences · doi:10.1145/3776759.3776861 · Cited by 2 · cited 4×
“Multi-agent smart-contract auditors [119,121–123 , Compared to coverage-guided fuzzers such as AFL++ and libFuzzer—which excel at exploring branch space within a single binary but treat every input byte as opaque—LLM-driven discovery contributes three capabilities.”
InfluentialHe, H., Yuan, X., Wu, K., & Ni, W. (2026) · Electronics · doi:10.3390/electronics15071409 · Cited by 2
iThelma: An Autonomous LLM Agent for Cyber Threat Hunting via Playbook-Driven Intelligence
Chen, N., Lin, R., Xie, D., Lin, H., & Chen, S. (2025) · 2025 IEEE Conference on Communications and Network Security (CNS) · doi:10.1109/cns66487.2025.11195050 · Cited by 2 · cited 8×
“SOAR-LLM [28], Agentic AI for threat hunting [24,25], and the SOC hyper-automation architecture of Alzughaibi et al. [28] follow this pattern.”
InfluentialWang, L., Sun, J., Jiang, J., Kanhere, S. S., Xing, Z., & Jha, S. K. (2025) · doi:10.1145/3709018.3736330 · Cited by 2
Wang, L., Sun, J., Jiang, J., et al. (2024) · doi:10.22541/au.173226080.05441422/v1 · Cited by 1
Lightweight Identification of Malicious IoT Traffic via Cross-View Knowledge Distillation
Nie, F., Liu, W., Liu, G., Gao, B., Huang, J., & Yuen, C. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3596167 · Cited by 1
LLM Assisted Dual-View Awareness Framework for Smart Contract Vulnerability Detection
Wang, J., Yue, Y., Hu, D., Li, Q., Li, J., & Zhu, W. (2025) · 2025 IEEE 36th International Symposium on Software Reliability Engineering (ISSRE) · doi:10.1109/issre66568.2025.00022 · Cited by 1 · cited 1×
“Multi-agent smart-contract auditors [119,121–123 , Compared to coverage-guided fuzzers such as AFL++ and libFuzzer—which excel at exploring branch space within a single binary but treat every input byte as opaque—LLM-driven discovery contributes three capabilities.”
Applying Multi-Agent LLMS to Threat Analysis and Risk Assessment in Automotive Security
Obradov, A., Pavković, B., Stojanović, D., & Četić, N. (2025) · 2025 33rd Telecommunications Forum (TELFOR) · doi:10.1109/telfor67910.2025.11314224 · Cited by 1 · cited 2×
“(ii) Debate coordination: agents argue from different positions to surface false negatives, as in debate-driven phishing detection [90] and the writer-reviewer automotive threat analysis loop [91].”
Huang, Z., Zhou, Z., Song, L., Deng, F., Zhang, L., SongGong, K., & Shen, M. (2026) · IEEE Internet of Things Journal · doi:10.1109/jiot.2026.3662325 · Cited by 1 · cited 2×
“Each layer exhibits distinct attack surfaces: at the perception layer, firmware exploits and command injection [40,41]; at the network layer, traffic flooding, protocol fuzzing, and lateral movement [42–45]; at the application layer, credential abuse, supply chain compromise, and data exfiltration…”
Chen, L., Zhao, C., Liu, Z., & Wang, H. (2024) · International Conference on Computer Science and Management Technology (ICCSMT) · doi:10.1145/3708036.3708092 · Cited by 1
Labeling NIDS Rules with MITRE ATT&CK Techniques: Machine Learning vs. Large Language ModelsPreprint
Daniel, N., Kaiser, F. K., Giladi, S., Sharabi, S., Moyal, R., Shpolyansky, S., Murillo, A., Elyashar, A., & Puzis, R. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2412.10978
Knowcti: Knowledge-Based Cyber Threat Intelligence Entity and Relation Extraction
Wang, G., Liu, P., Huang, J., et al. (2024) · doi:10.2139/ssrn.4711136
Research works printed but not linked: 6
- Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K.R., and Cao, Y. (2023, January 1–5). ReAct: Synergizing Reasoning and Acting in Language Models. Proceedings of the Eleventh International Conference on Learning Representations (ICLR 2023), Kigali, Rwanda. Available online: https://openreview.net/forum?id=WE_vluYUL-X.
- European Parliament and Council of the European Union (2026, May 29). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act). Official Journal of the European Union, L Series. Available online: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689.
- Deng, G., Liu, Y., Mayoral-Vilches, V., Liu, P., Li, Y., Xu, Y., Zhang, T., Liu, Y., Pinzger, M., and Rass, S. (2024). PentestGPT: Evaluating and Harnessing Large Language Models for Automated Penetration Testing. Proceedings of the 33rd USENIX Security Symposium (USENIX Security ’24), Philadelphia, PA, USA, 14–16 August 2024, USENIX Association. Available online: https://www.usenix.org/conference/usenixsecurity24/presentation/deng.
- Arp, D., Quiring, E., Pendlebury, F., Warnecke, A., Pierazzi, F., Wressnegger, C., Cavallaro, L., and Rieck, K. (2022). Dos and Don’ts of Machine Learning in Computer Security. Proceedings of the 31st USENIX Security Symposium (USENIX Security ’22), Boston, MA, USA, 10–12 August 2022, USENIX Association. Available online: https://www.usenix.org/conference/usenixsecurity22/presentation/arp.
- Judith, A., Kathrine, G.J.W., Silas, S., and Andrew, J. (2023). Efficient DL-Based Cyber-Attack Detection for IoMT Devices. Eng. Proc., 59.
- Kumar, P.V., Pittala, R.B., Jashwanth, P., Vedamsh, R., Chandrika, D.S., and Reddy, Y.V. (2025). Automated Threat Detection and Response Using SOAR and Endpoint Detection and Response. Proceedings of the 2025 OITS International Conference on Information Technology (OCIT), IEEE. Available online: https://ieeexplore.ieee.org/document/11400041/.
Cited by
Nageshwaran, V., & Ezekiel, S. (2026) · doi:10.20944/preprints202606.0142.v1 · Cited by 4
Self-citationAutonomous LLM Agent for Real-Time DDoS Attack Classification and Mitigation Rule Generation
Phani, K. V. S., Karthik, P. N., Mahammad, F. S., Subramanyam, M. V., & Srikanth, B. (2026) · International Conference on Intelligent and Sustainable AI Systems (ICOSAAS) · doi:10.1109/icosaas68663.2026.11648534 · Cited by 1
Garike, V. (2026) · Research Square · doi:10.21203/rs.3.rs-10582034/v1
Metadata
- Authors
- Nageshwaran, V., & Ezekiel, S.
- Affiliations
- University of the Cumberlands (US)
- Indiana University of Pennsylvania (US)
- Corresponding author
- Nageshwaran, Vinoth
- ORCID
- Nageshwaran, Vinoth
- Venue
- Electronics
- Publisher
- MDPI AG
- Published
- 2026-06-22
- Volume
- 15
- Issue
- 12
- Pages
- 2740
- ISSN
- 2079-9292 (ISSN-L)
- OpenAlex
- W7165680161
- Cited by
- 4
- References
- 128
- Citations by year
- 2026: 4
- FWCI
- 14.96
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Electronics (publishedVersion, cc-by)
- Preprints.org (acceptedVersion, cc-by)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Keywords
- agentic AI, large language models, Internet of Things security, multi-agent systems, chain-of-thought reasoning, React, tool use, threat detection and response, vulnerability detection, deception, edge computing, fog computing, cloud deployment, hallucinations, prompt injection, explainability, privacy, federated learning, research roadmap, taxonomy (general classification scheme)
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
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.