Attackers published two malicious versions, 1.26.1 and 1.26.2, of the agent-framework package on the Python Package Index on 17 November, 33 minutes apart. The package is the meta-package for Microsoft's Agent Framework, a Python library for building AI agents, and has about 190,000 downloads a month. Both versions have been quarantined, and version 1.26.0 is safe. It is unclear how the attacker got publishing access, and no Microsoft statement was included.
How the stealer worked
Each version installs a file that runs every time a Python interpreter starts, even if the package is never imported, so the stealer ran on any Python launch in an affected environment, including pip itself. It collected environment variables and credential files for Azure, npm, pip, Git, Docker, Hugging Face and Azure DevOps, Azure DevOps and managed-identity tokens, and cloud instance metadata, and requested fresh tokens from the platform, which rotating stored secrets would not cover. Version 1.26.2 also collected GitHub Actions tokens. The data was compressed, encoded and sent by HTTPS to a hard-coded webhook address.
What to do
Eight security firms reported it. Advice is to isolate affected hosts and build runners, pin to 1.26.0 and rebuild from a clean environment, check for exposure between about 8.35 pm and 9.20 pm UTC on 17 November, search proxy and DNS logs for requests to the webhook service, rotate exposed credentials from a clean system, and audit Azure identities and GitHub Actions workflows that deploy to Azure. It is the second compromise of a Microsoft-related package on the index this year, after three malicious versions of another Microsoft package in May. The report mentions no India or Southeast Asia link.










