Existing malicious code detection techniques demand the integration of multiple tools to detect different malware patterns, often suffering from high misclassification rates. Therefore, malicious code detection techniques could be enhanced by adopting advanced, more automated approaches to achieve high accuracy and a low misclassification rate. The goal of this study is to aid security analysts in detecting malicious packages by empirically studying the effectiveness of Large Language Models (LLMs) in detecting malicious code. We present SocketAI, a malicious code review workflow to detect malicious code. To evaluate the effectiveness SocketAI, we leverage a benchmark dataset of$5,115 \text{npm}$packages, of which 2,180 packages have malicious code. We conducted a baseline comparison of GPT3 and GPT-4 models with the state-of-the-art CodeQL static analysis tool, using 39 custom CodeQL rules developed in prior research to detect malicious Javascript code. We also compare the effectiveness of static analysis as a pre-screener with SocketAI workflow, measuring the number of files that need to be analyzed and the associated costs. Additionally, we performed a qualitative study to understand the types of malicious packages detected or missed by our workflow. Our baseline comparison demonstrates a 16 % and 9 % improvement over static analysis in precision and F1 scores, respectively. GPT-4 achieves higher accuracy with 99% precision and 97% F1 scores, while GPT-3 offers a more cost-effective balance at 91 % precision and 94 % F1 scores. Prescreening files with a static analyzer reduces the number of files requiring LLM analysis by$\mathbf{7 7. 9 \%}$and decreases costs by$\mathbf{6 0. 9 \%}$for GPT-3 and$\mathbf{7 6. 1 \%}$for GPT-4. Our qualitative analysis identified data theft, execution of arbitrary code, and suspicious domain categories as the top detected malicious packages.

The prevalent use of third-party components in modern software development, rapid modernization, and digitization have significantly amplified the risk of software supply chain attacks. Popular large registries like npm and PyPI are highly targeted malware distribution channels for attackers due to heavy growth and dependence on third-party components. Industry and academia are working towards building tools to detect malware in the software supply chain. However, a lack of benchmark datasets containing both malicious and neutral packages hampers the evaluation of the performance of these malware detection tools. The goal of our study is to aid researchers and tool developers in evaluating and improving malware detection tools by contributing a benchmark dataset built by systematically collecting malicious and neutral packages from the npm and PyPI ecosystems. We present MalwareBench, a labeled dataset of 20,792 packages (of which 6,659 are malicious) from the npm and PyPI ecosystems. We constructed the benchmark dataset by incorporating pre-existing malware datasets with the Socket internal benchmark data and including popular and newly released npm and PyPI packages. The ground truth labels of these neutral packages were determined using the Socket AI Scanner and manual inspection.