Attackers published two malicious versions of Microsoft Agent Framework, an open-source toolkit for building AI agents, on November 17. They went to PyPI, the main public repository for Python packages. According to BleepingComputer, versions 1.26.1 and 1.26.2 carried code that collected credentials for cloud and developer services, including Azure, AWS, GitHub, Hugging Face and OpenAI, and sent them to an attacker-controlled server. PyPI quarantined both releases shortly after upload. How the attacker gained publishing access is still unknown, and Microsoft had not commented by publication time.
The package has about 200,000 monthly downloads, so developer machines and build pipelines that pulled the bad versions automatically may be exposed. Security firms that tracked the attack advise rebuilding affected environments from the clean 1.26.0 release, rotating any reachable credentials and checking outbound network logs.
