Poisoned Microsoft AI Package Put Cloud Credentials at Risk

Poisoned Microsoft AI Package Put Cloud Credentials at Risk. Hackers managed to sneak two malicious versions of Microsoft's own Agent Framework package onto the official Python software repository this week.

This is a DailyCyber.News update, brought to you by BareMetalCyber.com, for Wednesday, November 18th, 2026.

Hackers managed to sneak two malicious versions of Microsoft's own Agent Framework package onto the official Python software repository this week. This is a tool developers use to build AI agents, and it gets about one hundred ninety thousand downloads a month, so the potential reach here was significant. The tricky part is that the tampered versions kept all the normal functionality working perfectly, while secretly running credential-stealing code in the background the moment Python started up, even if a developer never directly used the package. The malware went after cloud secrets, environment variables, and authentication tokens, specifically targeting Azure and GitHub Actions credentials. Both malicious versions have since been pulled from the repository, and Microsoft is investigating, but researchers say there's no trace of this malicious code in Microsoft's actual source repository, suggesting it was pushed out through some other channel entirely. If your organization uses this package, the priority right now is checking your development and CI/CD environments for the affected versions and rotating any credentials that might have been exposed. Don't wait for Microsoft's investigation to conclude before acting.

For the sources and the full Daily Cyber newsletter, visit DailyCyber.News.

Poisoned Microsoft AI Package Put Cloud Credentials at Risk
Broadcast by