Year: 2025
Type: article
Abstract: As AI code assistants become integral to developer workflows, they increasingly influence not just how software is written, but how it is architected.While existing studies have examined syntactic bugs and insecure code patterns produced by large language models (LLMs), little attention has been paid to the designlevel flaws these tools may introduce -flaws that impact authentication design, data flow, trust boundaries, and security architecture.This position paper explores the emerging class of AI-induced design flaws, presents real examples generated by mainstream tools, and outlines why these subtle risks are harder to detect than code-level issues.We argue for a new category of security analysis -design pattern drift in AI-assisted development -and call for automated security reviews that look beyond code correctness into architectural soundness.
Authors: Srajan Gupta, Anupam Mehta
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4412112303