🔍 Read the full analysis: Why The Microsoft AI Agent Framework PyPI Hijack Calls For A Response on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
The PyPI package for Microsoft’s Agent Framework had two malicious releases, versions 1.26.1 and 1.26.2, on November 17, 2026. Security advisories say the code could steal credentials when Python started in an affected environment; PyPI quarantined both versions, but users and organizations that installed them should treat those environments as exposed and rotate accessible secrets.
Two releases of Microsoft’s Agent Framework Python package on PyPI contained credential-stealing code, according to security advisories and researchers who analyzed the packages. Versions 1.26.1 and 1.26.2 were published on November 17 and quarantined by PyPI, but environments that installed either release may have exposed cloud, developer or CI credentials.
The GitHub Advisory Database says the malicious code was absent from the last clean release, version 1.26.0, and rates the incident critical. The affected package is named agent-framework, the official umbrella package for Microsoft’s open-source toolkit for building AI agents in Python. It is not described as a lookalike package. Microsoft’s installation instructions point users to it on PyPI, the public registry for Python software.
According to the advisory timeline cited by researchers, version 1.26.1 appeared at 20:35 UTC on November 17. PyPI quarantined it at about 21:06 UTC, according to StepSecurity. A second release, 1.26.2, followed at 21:08 UTC and was quarantined at 21:16 UTC. StepSecurity says the second upload came just over two minutes after the first release was quarantined. PyPI currently lists 1.26.0 as the latest version, released November 12.
The malicious code added a startup hook, so it could run when Python started in an affected environment without the user importing the framework, according to the advisory. It sought credentials and tokens accessible to developer machines and build environments, including cloud and AI service credentials. Researchers cited by the source material also reported that the second version added the ability to request GitHub Actions identity tokens. The framework continued to work, which could make the compromise less apparent to users.
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 Environments Face Greater Exposure
The incident matters because Python packages often run inside developer laptops, CI pipelines and build servers, where credentials can grant access to source code, cloud infrastructure and software releases. A compromise in an official package can reach users who followed ordinary installation instructions, rather than only people who downloaded an obvious imitation.
The threat is not limited to credentials stored in files. The advisory says the code could inspect environment variables and other accessible credential sources, and request fresh tokens from certain cloud and pipeline identity services. The OpenSourceMalware research group, as quoted by The Register, said that the combination of collected credentials and active token requests is especially concerning in cloud build environments because short-lived tokens may still be useful if stolen promptly.
StepSecurity and JFrog both point to a separate response issue: removing or quarantining a malicious version does not, by itself, establish that the publisher’s credentials or access have been revoked. StepSecurity said quarantining the first release did not immediately revoke the attacker’s publishing access. The second upload, which followed shortly afterward, makes the containment and access-control questions relevant beyond this package.
Products Worth Considering

Amazon Kindle Scribe (16GB) with Premium Pen, Tungsten

Apple Watch SE 3 GPS + Cellular 40mm Smartwatch

Samsung Galaxy Tab A11+ Tablet (6GB RAM, 128GB Storage, Gray)
As an affiliate, we earn on qualifying purchases.
A Second Upload Followed Quarantine
Microsoft Agent Framework is an open-source toolkit for creating AI agents in Python and coordinating them in workflows. The agent-framework package acts as an umbrella that installs the framework’s core component and integrations. Researchers cited in the source material said releases they examined from 1.21.0 through 1.26.0 had included source code, while 1.26.1 was uploaded as a wheel only. That difference was among the details examined during the investigation.
Several security firms, including Socket, StepSecurity, Wiz, JFrog and Endor Labs, analyzed or reported the malicious releases. The OpenSSF’s open-source malware database listed the case as MAL-2026-20101 on November 17, and the GitHub Advisory Database published advisory GHSA-764m-3ppc-5m73 early on November 18. These advisories provide technical details for defenders; this report does not reproduce indicators of compromise or attack instructions.
Microsoft’s response reported by The Register was limited: “We have made some repositories private while we investigate malicious content.” The source material said it had found no Microsoft security advisory at the time it was written. Wiz said it promptly reported its findings to Microsoft on November 17.
“We have made some repositories private while we investigate malicious content.”
— Microsoft, in a statement to The Register
Victim Count and Credential Use Unknown
The available source material does not give a confirmed number of affected installations or victims. Wiz was quoted as estimating almost 200,000 downloads a month for the package, but that figure is not a count of downloads during the exposure window, nor does it show how many installations used the malicious releases.
It is also unclear whether stolen credentials were used to access accounts, repositories or cloud resources. The source material does not report confirmed downstream intrusions, identify the attacker, or establish how the publishing access was compromised. Researchers reported that the inspected code had no remote-code-loading second stage and that the malware was not persistent beyond the installed package; those findings describe the code examined, not the status of every potentially affected environment.
Microsoft’s public statement, as quoted by The Register, does not explain which repositories were made private or whether publisher credentials were revoked. Further details on the investigation and its scope were not included in the source material.
Rebuild, Rotate, and Await Findings
The GitHub Advisory Database advises treating any environment that installed 1.26.1 or 1.26.2 as exposed. Its guidance is to rebuild those environments from a clean base using the clean 1.26.0 release and rotate credentials the affected environment could access, including CI and GitHub Actions secrets. Organizations should also review their own logs and incident-response procedures; the specific technical indicators are available in the relevant advisories.
Users and security teams will need to watch for updates from Microsoft, PyPI and the researchers as they clarify the publishing-access issue, the reach of the incident and whether any credentials were used after theft. Until those findings are available, the number of affected users and any downstream compromise remain unconfirmed.
Key Questions
Which versions of the Microsoft Agent Framework package were affected?
The affected PyPI releases were agent-framework 1.26.1 and 1.26.2. The GitHub Advisory Database identifies 1.26.0 as the last clean release cited in its advisory.
How could the malicious code run?
The advisory says the releases added a Python startup hook. That meant the code could run when Python started in an affected environment, without importing agent_framework.
What should an organization do if it installed an affected version?
The advisory recommends treating the environment as exposed, rebuilding it from a clean base on version 1.26.0, and rotating credentials it could reach, including CI and GitHub Actions secrets.
How many users were affected?
No confirmed victim or installation count was provided in the source material. Wiz’s estimate of almost 200,000 monthly downloads is a general package-download figure, not a count of installations of the malicious releases.
Has Microsoft confirmed that attackers used stolen credentials?
The source material does not report confirmed downstream use of stolen credentials. Microsoft said it had made some repositories private while investigating, but the scope and outcome of that investigation were not specified.
Source: ThorstenMeyerAI.com
