🔍 Read the full analysis: Microsoft’s AI Agent Framework PyPI Incident: A Response Checklist on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
PyPI briefly hosted malicious versions 1.26.1 and 1.26.2 of Microsoft’s agent-framework package on November 17, 2026. Advisories say the releases could steal credentials when Python started in an affected environment; defenders should treat installations as exposed, rebuild from a clean version, and rotate accessible secrets.
Two malicious releases of Microsoft’s agent-framework Python package were available on PyPI for a short time on November 17, and security advisories say they contained a credential stealer that could run when Python started. The affected versions, 1.26.1 and 1.26.2, have been quarantined; teams that installed either should treat the environment and credentials it could access as exposed.
The GitHub Advisory Database identifies versions 1.26.1 and 1.26.2 as malicious and says the stealer was absent from the preceding clean release, 1.26.0. The advisory rates the incident critical. PyPI records and researcher timelines place the first upload at 20:35 UTC and the second at 21:08 UTC on November 17. StepSecurity reported that PyPI quarantined the first release at 21:06 UTC and the second at 21:16 UTC.
The advisory says the code added a startup hook, so it could run when Python launched without the application importing agent-framework. It describes collection of developer and cloud credentials, including tokens and secrets available to the affected environment. Researchers also reported requests for certain cloud and pipeline identity tokens. The package’s ordinary framework functions continued to work, meaning normal operation would not necessarily alert users to the malicious code.
Security firms including Socket, StepSecurity, Wiz, JFrog and Endor Labs analyzed the releases. Wiz said it promptly reported its findings to Microsoft on November 17. PyPI now lists version 1.26.0 as the current release, according to the supplied source material. The advisory’s response guidance is to rebuild affected environments from a clean base and rotate credentials they could access, including CI and GitHub Actions secrets.
agent-framework on PyPI · UTC
Two poisoned releases, one evening
The official package of Microsoft’s Agent Framework, step by step. Blue: what happened. Green: the clean version.
November 12 to 18, 2026
- Clean 1.26.0 publishedStill the current version on PyPI
- Malicious 1.26.1 appearsFramework works normally, a credential stealer runs whenever Python starts
- PyPI quarantines 1.26.1After about 31 minutes
- Attacker uploads 1.26.2Renamed files, also targets GitHub Actions tokens
- PyPI quarantines 1.26.2Eight minutes later
- Malware databases list itMAL-2026-20101 (OpenSSF), GHSA-764m-3ppc-5m73 (GitHub)
Sources: GitHub Advisory Database GHSA-764m-3ppc-5m73, Nov 18, 2026; OSV MAL-2026-20101, Nov 17, 2026; PyPI project data, checked Nov 19, 2026; StepSecurity analysis (quarantine times).
Graphic: Powered by Thorsten Meyer AI
Why Build Credentials Are at Risk
The incident matters because Python packages are often installed automatically in developer workstations, container builds and continuous-integration pipelines. A compromised dependency can therefore reach secrets used to publish software, access cloud accounts or connect to AI services. The advisory’s description of execution at Python startup means that simply not importing the framework may not have prevented exposure after installation.
The timing also highlights a response challenge: quarantining one release does not necessarily stop a publisher with compromised credentials from uploading another. StepSecurity said the first quarantine did not immediately revoke publishing access; JFrog similarly noted that quarantining a version does not, by itself, invalidate a compromised publishing credential. That account points to the need to check the full publication window and rotate relevant publishing credentials, not just remove a package version.
For organizations, the central question is not only whether the framework was used directly, but whether either release entered an environment where valuable secrets were present. Those secrets may need rotation even if there is no visible sign of misuse, because the source material does not establish that the malware caused an observable failure.
Products Worth Considering

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Black

Bose QuietComfort Wireless Noise-Cancelling Headphones, Twilight Blue

DJI Mini 4K Drone with 4K UHD Camera
As an affiliate, we earn on qualifying purchases.
How the Two Releases Appeared
Microsoft’s Agent Framework is an open-source toolkit for building AI agents and coordinating workflows in Python. The PyPI package agent-framework is an umbrella package that brings in the framework’s core package and integrations. The supplied source says Microsoft’s installation instructions pointed users to this official package, rather than an unrelated copycat.
The clean version 1.26.0 was released November 12. Five days later, 1.26.1 appeared as a wheel-only release, unlike the earlier releases examined by Endor Labs, which had also included source code. PyPI quarantined it, but StepSecurity says version 1.26.2 was uploaded about two minutes later, with renamed files and additional capabilities. PyPI quarantined that release eight minutes after its upload.
The OpenSSF malware database listed the incident as MAL-2026-20101, and the GitHub Advisory Database published entry GHSA-764m-3ppc-5m73. The package’s reported scale adds potential reach: Wiz said it receives almost 200,000 downloads a month. That figure is a reported monthly download count, not a tally of affected installations or confirmed victims.
“We have made some repositories private while we investigate malicious content.”
— Microsoft, in a statement to The Register
Victim Counts and Exposure Scope
The supplied reporting does not provide a confirmed victim count or establish how many installations occurred during the exposure window. Wiz’s monthly download estimate does not show how many downloads involved the malicious versions, and downloads do not necessarily equal unique users or affected systems.
It is also not clear from the available material whether stolen credentials were used after collection, which organizations were affected, or how many credentials the malware successfully obtained. Microsoft’s quoted statement says it is investigating malicious content, but the source reports no detailed Microsoft security advisory. Researchers’ findings that they did not identify a remote-code-loading second stage or persistence beyond the installed package are limited to the code they inspected; they do not establish whether stolen credentials were misused.
Containment and Investigation Steps
Organizations should check package records, lockfiles, build logs and container histories for installations of 1.26.1 or 1.26.2, including indirect or automated updates. If either version entered an environment, the advisory’s guidance is to isolate and rebuild it from a clean base using the clean 1.26.0 release, then rotate every credential that environment could reach. That includes cloud, source-control, package-registry and CI secrets, as applicable.
Security teams should also review access and audit logs for relevant services, while avoiding assumptions that no exposure occurred simply because the application kept working. The detailed indicators and technical detection guidance are in the referenced advisories, according to the source. Microsoft’s investigation and any further updates from the maintainers, PyPI or security researchers may clarify the publishing-access compromise, scope and potential misuse. Until then, teams should base their response on whether an affected version was installed and what secrets were available to it.
Key Questions
Which versions of agent-framework were affected?
The GitHub Advisory Database identifies versions 1.26.1 and 1.26.2 as malicious. The preceding release, 1.26.0, is identified as the last clean version in the supplied material.
Could the malicious code run if an application did not use the framework?
According to the advisory, the code used a Python startup hook and could run when Python started without importing agent-framework. An installation in an environment could therefore present risk even if the framework was not explicitly used by the application.
What should an organization do if it installed an affected release?
Treat the environment as exposed, rebuild from a clean base using version 1.26.0, and rotate credentials the environment could access, including relevant CI and GitHub Actions secrets. Review service logs for suspicious access as part of the investigation.
How many users or organizations were affected?
No confirmed victim count is provided in the source material. Wiz’s estimate of almost 200,000 downloads a month is a general package-download figure, not a count of downloads of the malicious versions or confirmed affected users.
Has Microsoft published a detailed security advisory?
The supplied source reports a short statement to The Register that Microsoft had made some repositories private while investigating malicious content, but says it had found no Microsoft security advisory at the time of publication.
Source: ThorstenMeyerAI.com
