Microsoft makes repos private after PyPI hijack
Poisoned agent-framework versions stole credentials
Security
Agent Framework 1.26.1 and 1.26.2 · credential stealer · now quarantined
Written with AI assistance, reviewed by the ACN desk.
“The malicious agent-framework shows how attackers are targeting the AI developer ecosystem and the CI pipelines behind it.”
Hackers have compromised Microsoft’s official Agent Framework package on PyPI, the Python package index. BleepingComputer reports that two malicious versions, 1.26.1 and 1.26.2, were published on November 17. Agent Framework is Microsoft’s open-source toolkit for building and coordinating AI agents.
StepSecurity’s analysis says version 1.26.1 was uploaded at 20:35 UTC and quarantined about 31 minutes later. Version 1.26.2 followed about 2 minutes after that quarantine. PyPI quarantined it at 21:16 UTC. Both versions install a startup hook, a file Python runs at launch. The stealer runs even if the framework is never imported.
The malware collects environment variables, credential files and cloud identity tokens. Its targets include Azure, AWS, GitHub, Docker and Hugging Face, BleepingComputer reports. Version 1.26.2 also requests GitHub Actions identity tokens. The stolen data goes to an attacker-controlled webhook.site endpoint.
It is unclear how the attacker gained publishing access. BleepingComputer says Microsoft had not responded to its questions by publication. The Hacker News reports that this is the second Microsoft-related PyPI package compromised this year. It says the package draws about 190,000 downloads a month. StepSecurity names version 1.26.0, published November 12, as the last clean release.
“The malicious agent-framework shows how attackers are targeting the AI developer ecosystem and the CI pipelines behind it.”
Sponsor
Spot an error? Email corrections@aicodingnews.si. Fixes are logged on the corrections page.
Sponsor
Sponsor
Poisoned agent-framework versions stole credentials
Attackers used the open-source tool Cairn
A synthetic internet for prompt-injection tests
Reuters cites internal posts; Meta has not responded
Sponsor
Sponsor