🔍 Read the full analysis: How To Respond To A PyPI Hijack Targeting Microsoft’s AI Agent Framework on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
PyPI versions 1.26.1 and 1.26.2 of Microsoft’s Agent Framework package contained credential-stealing code and were quarantined on November 17, according to security advisories and researchers. Anyone who installed either release should treat the environment as exposed, rebuild from a clean base and rotate credentials it could access. The number of affected installations and any confirmed misuse of stolen credentials have not been reported.
Two releases of Microsoft’s official Python package for its Agent Framework contained credential-stealing code, according to the GitHub Advisory Database and security researchers. PyPI quarantined versions 1.26.1 and 1.26.2 on November 17, but developers who installed either should treat their environments as exposed and replace credentials that those systems could access.
The affected package, agent-framework, is published on PyPI, the main public registry for Python software, and Microsoft’s installation instructions point users to it. The GitHub Advisory Database says both releases included a credential stealer absent from the last clean version, 1.26.0, and rates the incident critical. Wiz described the package as receiving almost 200,000 downloads a month, although the number of installs of the compromised releases is unknown.
Version 1.26.1 was published at 20:35 UTC on November 17 and version 1.26.2 at 21:08 UTC, according to the advisory timeline. StepSecurity says PyPI quarantined the first release at 21:06 UTC, then quarantined the second at 21:16. The second upload came just over two minutes after the first was quarantined. PyPI’s current listed version is 1.26.0, released November 12, according to the source report.
The advisory says the malicious code ran when Python started in an affected environment, even if the user did not import the framework. Researchers reported that it searched for credentials and tokens available to developer systems and build environments. StepSecurity said it found no remote-code-loading second stage in the code it examined; that finding does not establish whether credentials were taken or used.
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 Systems Face Greater Risk
The incident reaches beyond users who deliberately ran an AI agent. A Python package’s startup hook can execute during other work in the same environment, so installing the release may expose a developer laptop, CI runner, build server or container even when the framework itself appears to work normally.
Credentials on those systems can provide access to cloud services, source-code repositories, package registries and AI platforms. The GitHub advisory specifically urges rotation of credentials the environment could reach, including CI and GitHub Actions secrets. In cloud build environments, short-lived identity tokens may also matter: the source report says version 1.26.2 added requests for GitHub Actions identity tokens, while the advisory describes active token requests from cloud and pipeline identity services.
The episode also shows why removing one bad release may not stop further uploads if the publishing credential remains active. StepSecurity said quarantining version 1.26.1 did not immediately revoke the attacker’s publishing access; JFrog made a similar point. That is a security response issue distinct from removing the malicious package itself.
Products Worth Considering

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

Logitech MX Creative Console with 9 Customizable Keys, Graphite

Anker SOLIX C300 Portable Power Station, 288Wh
As an affiliate, we earn on qualifying purchases.
How the Two Uploads Unfolded
Microsoft’s Agent Framework is an open-source toolkit for building AI agents in Python and coordinating them in workflows. The PyPI package agent-framework serves as an umbrella package for the framework’s core code and integrations. The source report says releases examined by Endor Labs from versions 1.21.0 through 1.26.0 included source code, while 1.26.1 was uploaded as a wheel only.
After the November 17 uploads, the OpenSSF malware database listed the case as MAL-2026-20101; the GitHub Advisory Database later published alert GHSA-764m-3ppc-5m73. Socket, StepSecurity, Wiz, JFrog and Endor Labs were among the security firms that analyzed the releases, according to the source report. Wiz said it promptly reported its findings to Microsoft on November 17.
Microsoft gave The Register a brief statement: “We have made some repositories private while we investigate malicious content.” The source report said it had not found a Microsoft security advisory at publication time. PyPI quarantine prevents the releases from being installed from the registry, but it does not remove copies already installed in environments.
“The releases contained a credential stealer that is not present in the last clean release, 1.26.0.”
— GitHub Advisory Database
Victim Counts and Credential Use
The number of people or organizations that installed versions 1.26.1 or 1.26.2 has not been reported in the supplied material. Wiz’s estimate of almost 200,000 downloads a month refers to the package overall, not confirmed downloads of the malicious releases, and should not be read as a victim count.
It is also not clear whether stolen credentials were used to access accounts or services, whether any data was taken beyond the reported credential collection, or which specific publishing account or credential enabled the uploads. Researchers’ findings about the inspected code do not answer those questions. The source report says Microsoft was investigating, but gives no detailed account of its findings or response.
Rebuild, Rotate, and Await Findings
Organizations should check dependency records, lockfiles, build logs and installed environments for either affected version. The advisory’s guidance, as reported in the source material, is to treat any environment that installed 1.26.1 or 1.26.2 as exposed, rebuild it from a clean base using 1.26.0, and rotate every credential it could reach. That includes relevant cloud, repository, CI, package-registry and AI-service credentials. Teams should follow the linked security advisories for detailed investigation steps rather than relying on package removal alone.
Further updates depend on findings from Microsoft, PyPI and the researchers investigating the uploads. Readers should look for details on the affected publishing access, the scale of installations, and any evidence that collected credentials were used. The source material does not give a date for a fuller Microsoft advisory or a final incident report.
Key Questions
Which package versions were affected?
The affected PyPI releases were agent-framework 1.26.1 and 1.26.2. The last clean version identified by the GitHub Advisory Database is 1.26.0.
What should I do if I installed one of them?
Treat the environment as exposed. Rebuild it from a clean base using version 1.26.0 and rotate every credential the environment could access, including relevant CI secrets, as advised in the source material.
Does removing the package from my environment resolve the risk?
Not by itself. Credentials may already have been accessed, so the reported guidance is to rebuild from a clean base and rotate exposed credentials, not just uninstall the package.
How many users were affected?
No victim or installation count for the two malicious versions has been reported in the supplied material. Wiz’s estimate of almost 200,000 downloads a month is for the package overall, not for the affected releases.
Has Microsoft confirmed stolen credentials were used?
The source material does not report confirmed misuse of stolen credentials. Microsoft told The Register it had made some repositories private while investigating malicious content; further findings were not available in the report.
Source: ThorstenMeyerAI.com
