🔍 Read the full analysis: A PyPI Hijack Targeted Microsoft’s AI Agent Framework. What Now? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Two versions of Microsoft’s official Python package, agent-framework, briefly carried credential-stealing code on PyPI on November 17, 2026. PyPI quarantined both releases, but developers whose environments installed either version should treat them as exposed, rebuild from a clean version and rotate credentials the environments could access.
Two malicious versions of Microsoft’s official Python package for its Agent Framework were briefly available on PyPI on November 17, carrying code that sought developers’ cloud keys and other credentials, according to security advisories and researchers. PyPI quarantined both releases, but any laptop, build server, container or CI environment that installed version 1.26.1 or 1.26.2 should be treated as exposed, with credentials it could reach rotated.
The affected package is agent-framework, the PyPI package linked from Microsoft’s installation instructions. The GitHub Advisory Database says versions 1.26.1 and 1.26.2 contained a credential stealer absent from the last clean release, 1.26.0, and rates the incident critical. The package’s framework functionality continued to work, making the malicious behavior less likely to be obvious to someone who installed it.
According to the timeline compiled by StepSecurity, version 1.26.1 was published at 20:35 UTC and quarantined at 21:06 UTC on November 17. A second release, 1.26.2, appeared at 21:08 UTC and was quarantined at 21:16 UTC. StepSecurity says the second upload came about two minutes after the first release was quarantined. PyPI currently lists 1.26.0, released November 12, as the available version.
The advisory says a startup hook made the malicious code run when Python started in an affected environment, even if the user did not import the Agent Framework. Researchers reported collection of credentials and tokens accessible to developer machines and build environments, including cloud, source-control and AI-service credentials. StepSecurity found no remote-code-loading second stage in the code it examined; The Register reported researchers’ assessment that the malware did not persist beyond the installed package.
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 Risk
The incident matters beyond users actively building AI agents because the package’s startup hook could run in an environment where Python was launched after installation. That can include ordinary development work as well as automated builds. A compromised build environment may hold cloud credentials, CI secrets and service tokens with access to systems beyond the developer’s machine.
Researchers described both passive collection of credentials and requests for fresh identity tokens from some cloud and pipeline services. OpenSourceMalware, quoted by The Register, said that combination is especially risky in cloud build environments because short-lived tokens may still be useful if captured and sent out quickly. The practical concern is not limited to whether the malicious package remains installed: credentials it accessed may remain valid until revoked or rotated.
The quick second upload also shows a limitation of responding only to a single bad release. StepSecurity said quarantining 1.26.1 did not immediately revoke the attacker’s publishing access. JFrog similarly warned that removing one version does not, by itself, disable a compromised publishing credential. The timeline makes control of the publishing account or credentials a key part of the investigation, though the source material does not establish how access was obtained.
Products Worth Considering

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray

Sonos Five Wireless HiFi Speaker, Black

Samsung Galaxy Tab A11+ Tablet (6GB RAM, 128GB Storage, Gray)
As an affiliate, we earn on qualifying purchases.
How the Two Releases Appeared
Microsoft’s Agent Framework is an open-source toolkit for building Python-based AI agents and coordinating them in workflows. The PyPI package agent-framework is an umbrella package that brings in the framework’s core code and integrations. Its presence on Microsoft’s own installation instructions made the releases part of a normal installation path, rather than an unrelated package using a similar name.
The incident followed a clean release on November 12. Endor Labs said the releases it examined from versions 1.21.0 through 1.26.0 had included source code, while 1.26.1 was published as a wheel only, according to the advisory. After PyPI quarantined 1.26.1, the second version appeared with renamed files and added capabilities, StepSecurity reported.
The OpenSSF’s open-source 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. 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.
“The versions contained a credential stealer that is not present in the last clean release, 1.26.0.”
— GitHub Advisory Database
Victim Count and Access Route
No victim count is provided in the source material, and it is not clear how many environments installed either release during the short availability window. Wiz says the package receives almost 200,000 downloads a month, but that figure does not show how many downloads involved the malicious versions or how many installations were affected.
The public information cited here does not explain how the attacker obtained publishing access, whether any stolen credentials were used, or whether Microsoft has confirmed exposure of its own systems. The source reports that the attacker’s collection address received stolen data, but provides no verified account of what credentials were captured or what happened to them. Microsoft’s investigation is ongoing, and the status of the publishing credentials is not established in the material.
Rebuild, Rotate, and Monitor
Organizations should check package records and deployment logs to determine whether any environment installed agent-framework 1.26.1 or 1.26.2. The GitHub advisory recommends treating affected environments as exposed, rebuilding from a clean base with version 1.26.0, and rotating every credential the environment could access, including CI and GitHub Actions secrets. Teams should also review relevant cloud and source-control activity for suspicious use of exposed credentials.
Developers should follow updates from Microsoft, PyPI and the published security advisories for information about the investigation and any revised remediation guidance. The source material does not give a date for a further Microsoft report or state when PyPI’s investigation will conclude. The key next step for affected organizations is to establish whether they installed either version and act on credentials accordingly, rather than relying only on the package’s removal from the registry.
Key Questions
Which package versions were affected?
The GitHub Advisory Database identifies agent-framework 1.26.1 and 1.26.2 as malicious. It names 1.26.0, released November 12, as the last clean release in this incident.
Could the code run if I did not import the framework?
Yes. The advisory says a startup hook ran when Python started in an affected environment, without requiring the user to import agent_framework.
What should an organization do if it installed an affected version?
Treat the environment as exposed, rebuild it from a clean base using version 1.26.0, and rotate credentials it could access. The advisory specifically includes CI and GitHub Actions secrets.
How many users were affected?
The source material gives no confirmed victim count. Wiz’s estimate of almost 200,000 downloads a month is a monthly package-download figure, not a count of installations of the two malicious versions.
Has Microsoft explained how the package was compromised?
Not in the material provided. Microsoft told The Register it had made some repositories private while investigating malicious content, but the access route and the investigation’s findings remain unclear.
Source: ThorstenMeyerAI.com
