Privacy Risk Profiling in Claude Code–Based LLM Access Architectures – A Systematic Analysis
DOI:
https://doi.org/10.15837/ijccc.2026.6.7577Keywords:
LLM privacy,, Threat modeling,, GDPR,, NIS2,, self-hosted AI,, coding assistants,, privacy risk scoring., STRIDEAbstract
The proliferation of Large Language Model (LLM) coding assistants has created a fragmented ecosystem of access architectures, ranging from fully local deployments to cloud-based commercial solutions. Organizations face increasing regulatory pressure under GDPR Article 25 (privacy by design) and the NIS2 Directive to justify their data processing choices. This study provides the first systematic privacy risk profile of 20 LLM access configurations documented in a 21-entry ontology, comparing self-hosted solutions against standard commercial offerings.
We employed STRIDE threat modeling per configuration, network traffic analysis in controlled environments (Wireshark/tcpdump), and CVSS-adapted quantitative scoring (0–10 Privacy Risk Score). Regulatory compliance was mapped to GDPR Art. 25, NIS2 Directive, and EU data residency requirements.
Pure local solutions achieved a mean Privacy Risk Score of 0.8/10 (range 0–2), outperforming all commercial baselines (mean 7.3/10, range 6.5–8.3). Hybrid routing architectures, which route sensitive projects locally and other traffic to the cloud, scored 3.2/10 in mixed use, offering optimal trade-offs. Windows 7–compatible solutions (S1.5, S2.8, S2.9) introduce NIS2 non-compliance risks due to absent security updates; of these, only S1.5 also offers pure local privacy. Self-hosted LLM architectures significantly reduce privacy risk compared to commercial alternatives. Organizations handling sensitive code should prioritize pure local deployments. The ontology-driven methodology is reproducible and extensible.
References
European Parliament and Council of the European Union, "Article 25: Data protection by design and by default," Regulation (EU) 2016/679 (General Data Protection Regulation), 2016.
European Parliament and Council of the European Union, "Directive (EU) 2022/2555 on measures for a high common level of cybersecurity across the Union (NIS 2 Directive)," Official Journal of the European Union, vol. L 333, pp. 80-164, Dec. 2022.
Microsoft Corporation, "STRIDE Threat Modeling," Microsoft Security Engineering, 2006.
FIRST.Org, Inc., "Common Vulnerability Scoring System v3.1: Specification Document," 2019.
OpenRouter, "OpenRouter API Documentation," 2026. [Online]. Available: https://openrouter.ai/docs
Ollama, "Ollama v0.14.0 Release," GitHub, Jan. 2026.
BerriAI, "LiteLLM Gateway (Proxy)," GitHub, 2024.
UpGuard Research Team, "YOLO Mode: Hidden Risks in Claude Code Permissions," UpGuard Blog, Jan. 2026.
H. He, W. Xu, R. Liu, J. Tang, B. Li, and X. Lin, "ECCC: Edge Code Cloak Coder for Privacy Code Agent," in Proceedings of the 9th Widening NLP Workshop, Suzhou, China, Nov. 2025, pp. 65-74. https://doi.org/10.18653/v1/2025.winlp-main.14
G. Gerganov, "llama.cpp: LLM Inference in C/C++," GitHub, 2025.
Qwen Team, "Qwen2.5-Coder-7B-Instruct-GGUF," Hugging Face, 2025.
Groq Inc., "GroqCloud API Documentation," Groq Console, 2026.
NVIDIA Corporation, "NVIDIA NIM Microservices," NVIDIA Developer, 2026.
Google, "Google Cloud Free Program," Google Cloud, 2026.
Amazon Web Services, "AWS Activate for Startups," AWS, 2026.
Anthropic, "Claude Code CLI," npm, 2026.
musistudio, "Claude Code Router (ccr)," GitHub, 2026.
ENISA (European Union Agency for Cybersecurity), "Guidelines on Security Measures under the NIS 2 Directive," ENISA Publications, 2024.
IETF, "RFC 8996: Deprecating TLS 1.0 and TLS 1.1," Internet Engineering Task Force, Mar. 2021.
K. Greshake, S. Abdelnabi, S. Mishra, C. Endres, T. Holz, and M. Fritz, "Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection," in Proceedings of the 2023 ACM Workshop on Artificial Intelligence and Security, 2023, pp. 79-90. https://doi.org/10.1145/3605764.3623985
N. Carlini, M. Jagielski, C. A. Choquette-Choo, D. Paleka, W. Pearce, H. Anderson, A. Terzis, K. Thomas, and F. Tramèr, "Poisoning Web-Scale Training Datasets is Practical," in Proceedings of the 2024 IEEE Symposium on Security and Privacy (S&P), 2024, pp. 1161-1178. https://doi.org/10.1109/SP54263.2024.00179
A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, "Learning Transferable Visual Models From Natural Language Supervision," in Proceedings of the 38th International Conference on Machine Learning (ICML), 2021, pp. 8748-8763.
A. Grattafiori et al., "The Llama 3 Herd of Models," arXiv preprint arXiv:2407.21783, 2024.
A. Yang et al. (Qwen Team), "Qwen3 Technical Report," arXiv preprint arXiv:2505.09388, 2025.
A. Chowdhery et al., "PaLM: Scaling Language Modeling with Pathways," Journal of Machine Learning Research, vol. 24, no. 240, pp. 1-113, 2023.
A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, "Attention Is All You Need," in Advances in Neural Information Processing Systems 30 (NeurIPS 2017), 2017, pp. 5998-6008.
Additional Files
Published
Issue
Section
License
Copyright (c) 2026 Nicolaie Popescu Bodorin, Mariana Nagy, J.L. de Miranda
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
ONLINE OPEN ACCES: Acces to full text of each article and each issue are allowed for free in respect of Attribution-NonCommercial 4.0 International (CC BY-NC 4.0.
You are free to:
-Share: copy and redistribute the material in any medium or format;
-Adapt: remix, transform, and build upon the material.
The licensor cannot revoke these freedoms as long as you follow the license terms.
DISCLAIMER: The author(s) of each article appearing in International Journal of Computers Communications & Control is/are solely responsible for the content thereof; the publication of an article shall not constitute or be deemed to constitute any representation by the Editors or Agora University Press that the data presented therein are original, correct or sufficient to support the conclusions reached or that the experiment design or methodology is adequate.
