2026, Vol.22, No.2

Journal of Software Forensics 2024 KCI Impact Factor : 0.32

pISSN : 3092-541X / eISSN : 3140-071X
https://journal.kci.go.kr/ksavs

Software completeness evaluation is important for software quality assurance, maintenance verification, outsourced software inspection, and public software project assessment. Existing approaches have limitations in evaluating whether software implementations semantically satisfy user requirements. This paper proposes an LLM-based software appraisal model that evaluates software completeness based on semantic alignment between requirements and source code. The proposed model extracts functional information from natural language requirements, analyzes semantic information from source code, and evaluates requirement-implementation alignment using embedding-based similarity and LLM-based reasoning. In addition, the model detects missing and over-implemented functionalities and automatically generates explainable evaluation reports. Experimental results show that the proposed model outperforms conventional keyword-based traceability approaches and improves the explainability and automation of software completeness evaluation. The proposed approach can be applied to software auditing, maintenance evaluation, outsourced software verification, and public software quality assessment.

The proliferation of AI coding tools poses a new challenge for determining copyright infringement in computer programs, particularly at the filtration step of the substantial similarity analysis. Because AI-generated code lacks human creativity, it is not protectable and is filtered out; yet in practice, human-authored and AI-generated portions are so intermingled that separating them is exceedingly difficult. The article finds the resulting problems fall into two categories: those that exacerbate existing issues and those specific to artificial intelligence. To address them, it proposes reinforcing the conventional filtration process, recalibrating the concept of reliance, and broadening the scope of comparison.

This paper examines the impact of the rapid spread of large language models (LLMs) and agent-based systems on software development methodologies and software appraisal systems, and proposes a new appraisal paradigm for the age of artificial intelligence. As software development shifts from explicit human design and implementation toward AI-collaborative development characterized by natural language-based requirement specification, AI-assisted code generation, and iterative cycles of planning, execution, and verification, software appraisal must also expand beyond completed source code and functional deliverables. It should encompass training data, model architectures and weights, prompts and generation histories, the traceability of human-AI collaboration, and external APIs and operational pipelines. From this perspective, this paper systematizes forms of AI utilization as “of the AI, by the AI, for the AI” and highlights the need for an appraisal framework covering AI models, AI-generated outputs and processes, and the ecosystem in which AI operates.