Observed incident

Two Microsoft Agent Framework releases on PyPI carried credential stealers

Reported

Published here

What happened

Malicious agent-framework 1.26.1 and 1.26.2 releases installed credential-stealing Python startup hooks. Wiz’s November 18 update links the operation to tooling identifying itself as Cairn and a captured Pi agent configuration referencing a GLM-family model.

Parties and reporting

Where this happened

Victim location not disclosed. Real organizations or systems were affected or targeted, but the sources do not say where. Explore the world map.

AI tools and systems: Microsoft agent-framework 1.26.1 and 1.26.2, Python .pth startup hooks, Cairn (tooling identification reported by Wiz), Pi / GLM-family configuration (reported by Wiz)

Outcome

PyPI quarantined both releases. Wiz analyzed 50 stored webhook requests: four from the attacker and 29 apparently from sandboxes, scanners or tests. It attributed the remaining 17 requests to three unique successful victim-data exfiltration cases. Those cases are not a complete victim census.

Operating lesson

Isolate affected environments before starting Python or pip, rebuild from a clean image, and investigate and rotate credentials the payload could access.

Evidence and limits

Wiz assesses apparent compromise of a Microsoft ESRP client certificate and private key; their acquisition remains unknown. Framework identifiers and captured configuration support AI-tool involvement, not full autonomy or attribution to a named operator. Three cases describe Wiz’s observed request set; they do not establish every installation or token request succeeded. First disclosure remains November 17.

Sources