AIShield – A Framework for QA in Software Development to Detect AI Generated Code

The rapid adoption of AI-generated code introduces new challenges in software Quality assurance (QA). Current static analysis tools fail to address AI-specific anti- patterns such as hallucinated APIs, contextual discontinuities, and architectural inconsistencies. This paper presents AIShield Sentinel, a novel validation framework that combines abstract syntax tree (AST) analysis with pattern-based detection to identify and quantify risks unique to AI-generated code. The system introduces three key innovations: (1) a context cohesion metric (0–100 scale) to measure logical flow between code segments, (2) architectural smell detection for AI-induced design flaws, and (3) an interpretable AI-probability score (0– 100%) based on weighted pattern matching. Evaluated against 1,200 code samples (GPT-4, Claude, and human-written), AIShield Sentinel achieves 92.3% precision in detecting AI- generated code artifacts—a 41% improvement over Pylint. The tool’s structured JSON output enables seamless integration with CI/CD pipelines while providing research- grade metrics for AI code quality analysis. This work bridges a critical gap in software development by offering the specialized validation framework for AI-assisted development environments.