Shai-Hulud Themed Malware Found in the PyTorch Lightning AI Training Library (semgrep.dev)

A recent supply chain attack has compromised the PyTorch Lightning AI training library, specifically affecting versions 2.6.2 and 2.6.3 released on April 30, 2026. This attack, linked to the infamous "Mini Shai-Hulud" campaign, involves a hidden JavaScript payload that executes upon importing the library. The malware is designed to steal sensitive credentials such as authentication tokens and environment variables while also attempting to disseminate itself via npm, leading to broader vulnerabilities across dependent packages and repositories. The significance of this incident lies in its implications for the AI/ML community, given that many projects rely on PyTorch Lightning for their model training processes. Notably, the malware not only targets local development environments but also infiltrates CI/CD pipelines and major cloud providers (AWS, Azure, GCP). This scenario illustrates a sophisticated level of persistence, utilizing hooks within popular developer tools like Claude Code and VS Code to ensure the payload remains active even with user intervention. Affected developers are urged to audit their repositories for indicators of compromise and rotate any exposed credentials immediately to mitigate potential damage.