The Microsoft Agent Framework, an open-source Python toolkit for AI development, was compromised on the PyPI repository. Attackers hijacked the package to distribute malware designed to harvest sensitive credentials and cloud platform tokens from developers.
The first malicious version, 1.26.1, was uploaded on November 17 and remained active for 31 minutes before removal. This version executed malicious code automatically upon Python startup, regardless of whether the framework was imported into the project.
A second attack followed shortly after, with version 1.26.2 appearing for 8 minutes. This iteration included enhanced capabilities to request GitHub Actions OIDC tokens and scan GitHub Runner directories for additional data.
The malware targeted credentials for services including Azure, AWS, GitHub, Docker, Hugging Face, and OpenAI. It also attempted to access local configuration files and monitor other active system processes.
Developers who downloaded versions 1.26.1 or 1.26.2 are advised to delete the files immediately. Security experts recommend revoking existing credentials and auditing cloud environments for unauthorized access.




