Python agent-framework-malware: 150,000+ developers at risk, credentials Azure/GitHub stolen
17 November 2026, developers around the world have downloaded unintentionally malware.
Versions 1.26.1 and 1.26.2 package agent-framework in PyPI were published in malicious activity.
33 minutes separated the two versions.
And each one contained credential stealer.
What happened
Malware
FACT (according to SITG Consulting):
Malicious versions of agent-framework were published on PyPI (Python Package Index).
Version: 1.26.1 and 1.26.2
Time between posts: 33 minutes
What steals
The malware is designed to steal credentials when Python starts:
- Credentials of Azure
- Credentials of Azure DevOps
- Credentials GitHub Actions
It runs automatically each time Python starts.
The scope
SPECULATION CONTROLLED (PyPI not released exact numbers):
You probably downloaded by:
- 150,000+ individual developers
- Thousands of CI/CD pipelines
- Companies that use agent-framework
Every download = compromised machine.
Why it was possible to
Vulnerability 1: blind Trust in PyPI
Developers assume that the packages on PyPI are legitimate.
There is No validation pre-download of code.
Vulnerability 2: PyPI does not validate code
PyPI is not running a malware scan in packages before you publish.
Publish is instant. Validation is manual/a post-hoc.
Vulnerability 3: Automatic execution
If agent-framework has the code in __init__.py running to the import:
- Malware is executed immediately
- In CI/CD pipelines = credentials stolen
- On developer laptops = credentials stolen
Preventive measures of monitoring
So that developers and companies should have:
1. Validation packages before use
- Scan of malware before adding to requirements.txt
- Check the source code of the package
- Compared with previous version
2. Monitoring CI/CD pipelines
- What’s new version of the package behaves differently?
- What New connections to locations unknown?
- Does New environment variables being accessadas?
3. Validation of credentials
- Detect when credentials Azure/GitHub are accessed
- Monitor export of secrets
- Correlate with package updates
4. Scan of the supply chain
- Monitor changes in dependencies
- Alert if dependency major is updated
- Validate changes to the code in dependencies
How to 24Cevent would have helped
Step 1: Detection on developer laptops
When malware is running (version 1.26.1 downloaded):
- 24Cevent: ‘Process Python accessing credentials Azure’
- Correlation: ‘Credentials being exfiltradas to unknown location’
- Action: Blocking of export credentials
Credentials are not stolen.
Step 2: Detection in CI/CD
In pipelines:
- 24Cevent: ‘GitHub Actions secrets being accessed by a process unknown’
- Correlation: ‘Secrets being sent out of the pipeline’
- Action: Blocking of communication + kill the process
Build fails safe, and expose secrets.
Measure 3: Detection of the supply chain
In the repository:
- Monitoring: ‘the Dependency agent-framework was updated’
- Scan: ‘New version has suspicious code’
- Correlation: ‘Code export credentials’
- Action: Revert to previous version + alert
Malware never comes in pipeline.
Step 4: Post-breach recovery
If credentials were stolen:
- 24Cevent automates: Rotation of stolen credentials
- Invalidation of tokens Azure
- Revocation of GitHub tokens
- Audit of what was agreed with these credentials
Lesson for developers
Developer supply chain is critical because:
- Developers have maximum permissions on your laptops
- Developers have access to all the secrets (Azure, GitHub, AWS)
- Developers rely on packages public without validating
A malware in popular package = mass exfiltration of secrets, corporate.
SOURCES CHECKED: