EUVIMEDSkopri EUVIMED
Open access · CC BY 4.0

Evidence-Aware LLM-Assisted Autonomous Network Security and Defence

Rodrigo Moutinho, José Silvestre Silva, Hélder Figueiredo Reia, Miguel Correia

Computers · 2026-11-15 · 15 (11) · 837

DOI: 10.3390/computers15110837 ↗

At a glance

Journal article in Computers, published online on 2026-11-15. Written by Rodrigo Moutinho and 3 co-authors. The full text is freely available (open access, CC BY 4.0). Funded by: Centro de Investigação Desenvolvimento e Inovação da Academia Militar.

AstrattDettalji biblijografiċiCiteRelated studies

Abstract (original language)

Automated network response must stop attacks without restricting legitimate hosts. We evaluate deterministic evidence checks applied before a source is blocked, in a pipeline that combines XGBoost flow scoring, optional arbitration by a local language model (LLM) or by rules derived from its prompt, and SDN enforcement. All main comparisons were run live with an agent corrected for residual records between runs (504 runs over 14 attack scenarios). With the LLM held fixed, the checks cut benign blocks from 199 to 12, but lost the detections of two scenarios in which a wrong attack label selected a check. Without an LLM, they cut benign blocks from 191 to 11, detecting 27 instead of 30 of 42 runs. Almost all of this reduction came from one label-free check of destination dispersion. Stricter score thresholds traded detections for blocks along a curve that the checks exceeded only slightly. Prompt-derived rules matched or exceeded the LLM on every outcome, and injected text made the deployed model block a victim profile. Detection differences were not significant at the scenario level, and the thresholds and prompt were informed by the same testbed traffic. The results favour explicit evidence checks over LLM arbitration.

Funding

  • Centro de Investigação Desenvolvimento e Inovação da Academia Militar

Cite

APA

Moutinho, R., Silva, J. S., Reia, H. F., & Correia, M. (2026). Evidence-Aware LLM-Assisted Autonomous Network Security and Defence. Computers, 15(11), 837. https://doi.org/10.3390/computers15110837

Vancouver

Moutinho R, Silva JS, Reia HF, Correia M. Evidence-Aware LLM-Assisted Autonomous Network Security and Defence. Computers. 2026;15(11):837. doi:10.3390/computers15110837

BibTeX
@article{Moutinho2026,
  author = {Moutinho, Rodrigo and Silva, José Silvestre and Reia, Hélder Figueiredo and Correia, Miguel},
  title = {Evidence-Aware LLM-Assisted Autonomous Network Security and Defence},
  journal = {Computers},
  year = {2026},
  volume = {15},
  number = {11},
  pages = {837},
  doi = {10.3390/computers15110837}
}
Start for free →