Executive Summary
Socket Threat Research has identified a significant expansion of the Mini Shai-Hulud/Miasma campaign, involving 23 newly discovered malicious PyPI packages. This wave specifically targets the bioinformatics research community and developers working with the Model Context Protocol (MCP) and AI orchestration tools like LangChain. The attackers have moved beyond simple .pth hooks, now utilizing trojanized native extensions (.abi3.so) and decoupled loader-payload architectures to evade traditional static analysis.
The campaign's primary objective is the exfiltration of high-value secrets from developer workstations and CI/CD environments. Targets include cloud credentials, SSH keys, GitHub/npm/PyPI tokens, and Kubernetes service accounts. The malware also employs a unique anti-analysis technique by including a large JavaScript comment header designed to trigger safety refusals or context pollution in LLM-based security scanners, potentially blinding AI-driven triage tools to the underlying Hades obfuscated payload.
This evolving campaign demonstrates high agility, with the threat actors rapidly iterating through delivery mechanisms. The inclusion of reputable scientific research packages like 'embiggen' and 'ensmallen' indicates a focused effort to compromise high-value data in the scientific and AI sectors. Organizations using these packages or maintaining AI/bioinformatics development environments should perform immediate impact assessments.