Compromised Microsoft Agent Framework PyPI Releases Target Developer Credentials

Summary
Two malicious Microsoft Agent Framework releases on PyPI ran during Python startup to collect developers’ credentials and other sensitive data. PyPI quarantined both; affected developers are advised to rebuild environments and rotate exposed credentials.
Key points
- Malicious versions 1.26.1 and 1.26.2 were published on November 17, 2026, under Microsoft's official Agent Framework PyPI project.
- The credential stealer ran automatically during Python startup, even if the Agent Framework package was not imported.
- Version 1.26.1 targeted secrets in environment variables, configuration files, and accessible processes, including credentials for cloud and developer services.
- Version 1.26.2 added GitHub Actions OIDC token requests and searches of GitHub runner directories.
- The releases used different .pth-based startup mechanisms and sent collected data to the same attacker-controlled endpoint.
- PyPI quarantined both releases; how the attacker obtained publishing access remains unclear.
- Developers who installed either version should isolate affected environments, rebuild using the safe 1.26.0 release, and rotate potentially exposed credentials.
Article Details
- Victim Organization
- Microsoft
- Incident Type
- Compromise of Microsoft’s PyPI package with malicious releases designed to steal credentials
- Incident Date
- 2026-11-17
- Disclosure Date
- 2026-11-18
- Data Types Exposed
- Secrets from environment variables and local configuration files
- Azure managed identity access tokens and Azure DevOps OIDC tokens were targeted
- GitHub Actions OIDC tokens were requested by version 1.26.2
- Affected Records
- Unknown
- Affected Data Size
- Unknown
- Operational Impact
- Malicious versions 1.26.1 and 1.26.2 executed during Python startup in affected environments and attempted to collect and exfiltrate credentials and other sensitive data. PyPI quarantined both releases. The article advises users who installed either version to isolate and rebuild affected environments and rotate potentially exposed credentials; it does not establish how many environments were affected or whether particular transfers completed.
- Ransom Or Extortion
- Unknown
- Claim Status
- confirmed
Indicators of compromise
| Type | Indicator | Context |
|---|---|---|
| URL | hxxps[:]//webhook[.]site/d381fd80-ec62-42ad-8f5c-d1b15103c9bb | Attacker-controlled endpoint identified as the destination used by both malicious package versions to exfiltrate collected data; completed transfers may vary. |
Vendors
Products
AWSlocal configuration files, and other running processes where permissions allowed, targeting Azure, AWS, GitHub, Docker, Hugging Face, OpenAI, and other services.Azurevariables, local configuration files, and other running processes where permissions allowed, targeting Azure, AWS, GitHub, Docker, Hugging Face, OpenAI, and other services.Dockerconfiguration files, and other running processes where permissions allowed, targeting Azure, AWS, GitHub, Docker, Hugging Face, OpenAI, and other services.GitHublocal configuration files, and other running processes where permissions allowed, targeting Azure, AWS, GitHub, Docker, Hugging Face, OpenAI, and other services.Hugging Facefiles, and other running processes where permissions allowed, targeting Azure, AWS, GitHub, Docker, Hugging Face, OpenAI, and other services.Microsoft Agent FrameworkMicrosoft Agent Framework is Microsoft's open-source toolkit for building AI agents and coordinating them in workflows.OpenAIother running processes where permissions allowed, targeting Azure, AWS, GitHub, Docker, Hugging Face, OpenAI, and other services.PyPIMicrosoft’s Agent Framework PyPI package compromised to steal credentials
