🤖 AI Summary
Hackers compromised Microsoft's official Agent Framework Python package on PyPI by releasing two malicious versions—1.26.1 and 1.26.2—on November 17, 2026. These versions, downloaded roughly 200,000 times a month, were designed to stealthily extract sensitive information, including cloud credentials from services like Azure, AWS, GitHub, and Docker, during standard Python startup. The malware executed from within the package's structure, allowing it to run unnoticed while the legitimate functionality remained intact.
This incident highlights significant vulnerabilities in how open-source packages are managed and published, raising concerns within the AI/ML community about the security of widely-used tools. The malicious code's ability to harvest sensitive tokens, such as Azure managed identity access tokens and GitHub Actions OIDC tokens, could lead to severe security breaches. Microsoft has not yet clarified how the attacker gained publishing access, prompting a call to action for developers to isolate affected environments, secure their systems, and immediately rotate any potentially exposed credentials.
Loading comments...
login to comment
loading comments...
no comments yet