Malware spreads through Microsoft’s Agent Framework
Two malicious Python versions attempted to steal cloud and developer data.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
Two versions of Microsoft’s Agent Framework on the PyPI software platform contained malware capable of collecting passwords and other secret data. The infected packages were published on 17 November and removed shortly afterwards.
The versions concerned are Python versions 1.26.1 and 1.26.2 of agent-framework. According to cybersecurity firm Wiz, those releases contained a so-called credential stealer. The code attempted to read environment variables, where developers often store tokens, passwords and API keys.
According to several researchers, the malware could run automatically when Python started. The package did not need to be actively called for this to happen. As a result, developers or automated build systems that had installed the framework could also be affected, even if they were not deliberately using the component.
Researchers from Wiz and other cybersecurity firms report that Azure, Azure DevOps and GitHub Actions data were the main targets. Such keys can provide access to cloud environments, source code or software that is built and deployed automatically. Whether data was actually stolen varies by affected environment and has not been established publicly.
According to cybersecurity publications, the infected versions were on PyPI for less than an hour. That limits the potential spread, but does not rule out damage: packages may have been downloaded automatically by development environments during that period. Organisations that used either version may therefore face a risk of key theft.
The case is an example of a supply-chain attack. The attacker does not directly target every victim, but attempts to infect software trusted by many developers. The risk is growing because AI agents often gain access to corporate data, development environments and cloud infrastructure.
For users, the key question is not whether Microsoft itself wrote the malware. The researchers describe an infected distribution of a package made available under the name of Microsoft’s framework. Companies that used the affected versions should check their logs and rotate tokens in line with their own security procedures; public sources do not provide a complete overview of all affected organisations.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The article’s technical core is supported by analyses from several cybersecurity firms. Unknown damage and victim numbers are explicitly marked as uncertain.
- confirmed Versions 1.26.1 and 1.26.2 contained a credential stealer — SecurityWeek, BleepingComputer and Wiz describe both versions. source
- confirmed The releases appeared on 17 November 2026 — Wiz gives the publication date and time. source
- confirmed The malware targeted tokens and cloud data — The researchers mention Azure, GitHub and other secrets in environment variables. source
- confirmed The number of organisations actually affected is unknown — The public analyses describe how it worked, but do not provide a complete victim register. source