TechFlow reports that on November 18, security research firms Wiz and OX Security discovered that versions 1.26.1 and 1.26.2 of Microsoft’s Python toolkit agent-framework, used for building AI agents, contained credential-stealing code, and PyPI has isolated these two versions.
Wiz stated that after installing the affected versions, the malicious code can run upon the next Python startup, even if the application did not import the package. Researchers found that the code collects credentials such as environment variables, Azure CLI, and Git, and attempts to send them to attacker-controlled endpoints.
Both organizations advised affected developers to remove the relevant versions and rotate exposed credentials. Wiz stated that it remains unconfirmed how the attackers obtained publishing permissions.





