Logo image
Back
A Survey on the Role of LLMs in AI-Based Software Development: Augmentation and Latent Risks
Conference proceeding   Peer reviewed

A Survey on the Role of LLMs in AI-Based Software Development: Augmentation and Latent Risks

Md Bajlur Rashid, Mohammad Shafayet Jamil Hossain, Mohammad Ishtiaque Khan, Sharaban Tahora, Aiasha Siddika, Mahmudul Islam Prakash, Sharmin Yeasmin and Hossain Shahriar
IEEE International Conference on Big Data, (2025), pp.4343-4352
IEEE International Conference on Big Data (BigData) (Macau, China, 12/08/2025–12/11/2025)
12/08/2025
DOI: https://doi.org/10.1109/BigData66926.2025.11401846

Metrics

1 Record Views

Abstract

Large Language Models (LLMs) such as GPT-4, Codex, Code Llama, Claude, and DeepSeek are increasingly shaping AI-based software development. Their role is inherently dual: on one side, LLMs augment established practices by enabling faster vulnerability detection, supporting secure code generation, and assisting in continuous threat modeling; on the other, they introduce latent risks through insecure code suggestions, data leakage, and adversarial misuse. This survey examines recent studies from 2020-2025 to assess how LLMs enhance traditional frameworks such as the Secure Development Lifecycle (SDL) and DevSecOps, while also identifying emerging risks that threaten long-term reliability and compliance. Case studies reveal both the acceleration of secure coding practices and the unintended propagation of unsafe patterns. Further, challenges such as dataset contamination, probabilistic outputs, scalability issues, and developer over-reliance amplify these risks. By consolidating empirical findings, benchmarking studies, and sector-specific applications, this survey provides a structured view of the opportunities and vulnerabilities posed by LLMs in AI-driven software engineering, underscoring the need for governance, dataset curation, and hybrid human-AI collaboration to ensure trustworthy adoption.

Details

Logo image