🔍 Read the full analysis: The AI Agent Framework Hijack On PyPI: Who Needs To Take Action? on ThorstenMeyerAI.com
Get the latest gadgets 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
PyPI briefly hosted malicious versions 1.26.1 and 1.26.2 of Microsoft’s agent-framework package on November 17, 2026. Security advisories say the releases contained credential-stealing code; anyone who installed either should treat the environment as exposed, rebuild from a clean base and rotate credentials it could access.
Two releases of Microsoft’s agent-framework Python package distributed through PyPI contained credential-stealing code, according to the GitHub Advisory Database and security researchers. The versions, 1.26.1 and 1.26.2, were published on November 17 and quarantined within minutes; developers and organizations that installed either should treat the affected environment as exposed and rotate credentials it could reach.
The GitHub Advisory Database says the malicious code was absent from the last clean release, version 1.26.0, published November 12. PyPI records place version 1.26.1 at 20:35 UTC on November 17. StepSecurity says PyPI quarantined it at 21:06 UTC. A second release, 1.26.2, was uploaded at 21:08 UTC and quarantined at 21:16 UTC, according to the same timeline. Both releases are no longer available from PyPI; the source says version 1.26.0 is the current release there.
The advisory says the code ran when Python started in an affected environment, even if the user did not import the framework. Researchers report that it sought credentials and tokens associated with cloud services, developer tools, package registries and AI services. The advisory recommends treating an environment that installed either version as exposed, rebuilding it from a clean base using 1.26.0, and rotating credentials it could access, including CI and GitHub Actions secrets.
Microsoft’s Agent Framework is an open-source toolkit for building and coordinating AI agents. The agent-framework package on PyPI is an umbrella package that pulls in the core framework and integrations. The supplied source says Wiz estimated almost 200,000 downloads a month, but does not provide a period or method for that figure. Download volume does not establish how many installations received the malicious releases.
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
Credential Risk Across Build Systems
The incident puts more than a developer’s local machine at potential risk. Build servers, CI pipelines and containers can hold credentials that grant access to source repositories, cloud accounts, deployment systems or AI services. If malicious code ran in an environment with those secrets available, rotating passwords alone may not address exposure of tokens or other credentials.
The releases reportedly left the framework working normally, so routine use may not have revealed the compromise. The advisory says the code could execute on later Python starts without an explicit import of the package. That makes inventory and review of installed versions important, including in automated builds and images created during the affected period. These are precautionary steps; the supplied material does not establish that every installation resulted in stolen credentials.
The sequence also highlights a publishing-access issue. StepSecurity says quarantining the first release did not immediately revoke the publisher’s access, and the attacker uploaded another version shortly afterward. JFrog likewise says quarantining a release does not itself invalidate a compromised publishing credential. The account or credential used to publish the packages is therefore a separate part of the response from removing the files.
Products Worth Considering

DJI Mini 4K Drone with 4K UHD Camera, Under 249 g

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Bose SoundTouch 30 Wireless Speaker, Black
As an affiliate, we earn on qualifying purchases.
How the Releases Reached PyPI
The timeline in the supplied advisories begins with Microsoft’s legitimate 1.26.0 release on November 12. Five days later, 1.26.1 appeared on PyPI as a wheel-only release, according to the GitHub advisory. StepSecurity places its quarantine at 21:06 UTC. Version 1.26.2 followed at 21:08 UTC, with renamed files and additional capabilities described by researchers, and was quarantined at 21:16 UTC.
The OpenSSF malware database listed the case as MAL-2026-20101 on November 17, and the GitHub Advisory Database published GHSA-764m-3ppc-5m73 early on November 18, according to the source article. Security companies named in the material include Socket, StepSecurity, Wiz, JFrog and Endor Labs. Their findings and timelines provide the documented basis for the incident details here; the supplied material does not include a Microsoft security advisory.
“We have made some repositories private while we investigate malicious content.”
— Microsoft, in a statement to The Register
Victim Count and Exposure Scope
The supplied source does not give a confirmed number of affected installations or victims. It also does not establish how many users installed 1.26.1 or 1.26.2, whether credentials were successfully taken from particular environments, or whether any stolen credentials were used afterward. A package download estimate is not a count of compromised systems.
Microsoft’s brief statement says repositories were made private during an investigation, but the source reports no detailed Microsoft security advisory. It remains unclear which publishing credential or account was compromised, how access was obtained, and whether other packages or systems were affected. Researchers’ analysis that they found no remote-code-loading second stage or persistence beyond the installed package is limited to the code they inspected and does not resolve the broader questions.
Rebuild, Rotate, and Verify
Organizations should check package inventories, lockfiles, build logs and container histories for versions 1.26.1 and 1.26.2. If either was installed, follow the advisory’s guidance: isolate and rebuild the environment from a clean base on version 1.26.0, then rotate every credential the environment could access. Teams should include cloud, source-control, CI, package-registry and AI-service credentials in that review, and check relevant access records for suspicious use.
Security teams should follow the GitHub and OpenSSF advisories for technical indicators and updates; this report does not reproduce indicators or attack instructions. Further clarity depends on Microsoft’s investigation and any updates from PyPI, the advisory maintainers and the researchers. Until those details are published, the extent of exposure and any downstream use of credentials remain unconfirmed.
Key Questions
Which package versions were affected?
The reported malicious releases were agent-framework 1.26.1 and 1.26.2. The supplied source identifies 1.26.0, published November 12, as the last clean release.
Who should take action?
Anyone whose laptop, build server, CI pipeline or container installed either affected version should treat that environment as exposed. This includes installations caused by automated dependency updates.
What should an affected team do?
Rebuild the environment from a clean base using version 1.26.0, then rotate credentials the environment could reach, including relevant cloud, CI, GitHub Actions, package-registry and AI-service credentials.
How many users were compromised?
No confirmed victim or installation count is provided in the supplied source. The cited monthly download estimate is not a count of affected installations, and credential theft in any particular environment has not been established.
Has Microsoft explained how the releases were published?
The source reports a short Microsoft statement saying some repositories were made private while the company investigated malicious content. It does not provide a detailed account of the publishing compromise or a Microsoft security advisory.
Source: ThorstenMeyerAI.com
