EUVIMEDEUVIMED entdecken
Crossref

Trustworthy and Secure LLM-Assisted Code Generation for Enterprise Software Development

Sai Shiva Reddy Kongari, Saurav Kant Kumar, Abhishek Kumar

International journal of data science and machine learning · 25.05.2026 · 06 (01) · 222-237

DOI: 10.55640/ijdsml-06-01-04 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International journal of data science and machine learning, online erschienen am 25.05.2026. Verfasst von Sai Shiva Reddy Kongari und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large Language Models (LLMs) are increasingly embedded in enterprise software development workflows for source-code generation, debugging, documentation, test creation, refactoring, and design assistance. While these systems improve productivity and reduce repetitive programming effort, their deployment in enterprise environments introduces significant risks related to insecure code suggestions, hallucinated functionality, non-compliant logic, weak explainability, dependency misuse, and over-reliance by developers. This research paper investigates how LLM-assisted code generation can be made secure, reliable, and trustworthy through a structured governance-oriented framework that integrates prompt constraints, retrieval support, policy-based filtering, automated validation, secure coding guardrails, and human-in-the-loop review. Drawing only on the provided literature, the study positions enterprise code generation within the broader development of LLMs, code-specialized models, hardware generation models, design automation agents, and AI-assisted engineering workflows. Prior work on GPT-4, StarCoder, CodeGen2, Magicoder, VerilogEval, RTLLM, OpenLLM-RTL, ChipNeMo, ChipGPT, AutoChip, and secure hardware generation demonstrates that LLMs can generate technically meaningful artifacts but remain vulnerable to correctness gaps, evaluation instability, insufficient domain grounding, and security-sensitive errors (Achiam, 2023; Li, 2023; Nijkamp et al., 2023; Wei et al., 2023; Liu et al., 2023; Lu et al., 2023; Thakur et al., 2023). This paper extends those insights to enterprise software engineering and proposes a Trustworthy Secure Code Generation Framework organized around six layers: task specification, retrieval-augmented contextualization, constrained generation, security-policy enforcement, validation and testing, and accountable human approval. The findings suggest that trustworthiness in LLM-assisted enterprise code generation cannot be achieved by model scale alone. Instead, it requires layered controls that combine technical verification, organizational policy, developer accountability, and continuous feedback. The proposed framework reduces the probability of insecure code propagation, improves explainability of generated outputs, supports compliance-oriented review, and transforms LLMs from autonomous code producers into controlled development assistants. The paper concludes that enterprise adoption of LLM-assisted coding should prioritize measurable trust, secure-by-design workflows, traceable decision-making, and hybrid human-AI governance.

Zitieren

APA

Kongari, S. S. R., Kumar, S. K., & Kumar, A. (2026). Trustworthy and Secure LLM-Assisted Code Generation for Enterprise Software Development. International journal of data science and machine learning, 06(01), 222-237. https://doi.org/10.55640/ijdsml-06-01-04

Vancouver

Kongari SSR, Kumar SK, Kumar A. Trustworthy and Secure LLM-Assisted Code Generation for Enterprise Software Development. International journal of data science and machine learning. 2026;06(01):222-237. doi:10.55640/ijdsml-06-01-04

BibTeX
@article{Kongari2026,
  author = {Kongari, Sai Shiva Reddy and Kumar, Saurav Kant and Kumar, Abhishek},
  title = {Trustworthy and Secure LLM-Assisted Code Generation for Enterprise Software Development},
  journal = {International journal of data science and machine learning},
  year = {2026},
  volume = {06},
  number = {01},
  pages = {222--237},
  doi = {10.55640/ijdsml-06-01-04}
}
Kostenlos starten →