Malicious agent-framework PyPI Releases Target Developer and CI Credentials

· Original article ↗

Summary

OpenSourceMalware reports that agent-framework versions 1.26.1 and 1.26.2 contain a stealer that runs at Python startup, collects credentials and requests CI/cloud tokens, then sends data to a UUID-specific webhook.site URL.

Key points

  • The article identifies PyPI versions 1.26.1 and 1.26.2 as malicious releases published through Microsoft's PyPI publisher account.
  • A .pth file runs the stealer during ordinary Python startup without requiring the package to be imported; the article says this is not operating-system persistence.
  • The malware searches environment variables, credential files and other processes' environments for secrets.
  • It requests fresh GitHub Actions OIDC, Azure DevOps OIDC and Azure managed identity tokens when the relevant runtime endpoints and credentials are available.
  • Collected data is compressed, Base64-encoded and sent by HTTPS POST to a UUID-specific path on webhook[.]site.
  • The article recommends locating affected installs, rebuilding environments, rotating exposed credentials, revoking runtime credentials and reviewing Azure activity.

Article Details

Attack Vectors
  • Malicious agent-framework versions 1.26.1 and 1.26.2 were published through Microsoft's official PyPI publisher account.
  • A .pth file in the package imports af_runtime_diag during Python startup, causing the stealer to run without the package being imported.
  • The stealer collects secrets from environment variables, credential files, and other processes' environments, and requests fresh GitHub Actions, Azure DevOps, and Azure managed identity tokens.
  • Collected data is compressed, Base64-encoded, and sent by HTTPS POST to a UUID-specific webhook.site URL.
Defensive Notes
  • Treat installations of agent-framework 1.26.1 or 1.26.2 as credential compromises; isolate affected systems and stop compromised CI/CD jobs or runners.
  • Search dependency manifests, caches, build logs, and Python environments for the affected versions; inspect site-packages for af_runtime_diag.pth, af_runtime_diag.py, and unexpected executable .pth lines.
  • Rebuild affected environments from known-good dependencies and images, rotate exposed credentials, revoke GitHub Actions and Azure DevOps runtime credentials, and invalidate OIDC-derived sessions where possible.
  • Review Azure activity for use of managed identities and tokens tied to affected workloads, and check network telemetry for requests to the UUID-specific webhook path.

Indicators of compromise

TypeIndicatorContext
SHA256325e7e9f0ec94e41ad1a1b7a3f18a6b8b5bffece2e2b97b8d3369eb967452f93SHA-256 of af_runtime_diag.pth, the malicious Python startup file in the package.
SHA25654d885b7d7161c9ed96489c66364c4bfbf635929166843bba10d97d46b5a2e1fSHA-256 of af_runtime_diag.py, the credential-stealing module in the package.
URLhxxps[:]//webhook[.]site/d381fd80-ec62-42ad-8f5c-d1b15103c9bb?t=telemetrySpecific webhook.site URL identified as the HTTPS POST destination for exfiltrated data.

MITRE ATT&CK

Vendors

Products

Related Articles