🔍 Read the full analysis: How The PyPI Hijack Of Microsoft’s AI Agent Framework Put Credentials In Danger 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
Two versions of Microsoft’s Python package agent-framework, 1.26.1 and 1.26.2, were published with credential-stealing code on November 17, 2026. PyPI quarantined both releases, but anyone who installed either should treat the environment as exposed, rebuild from a clean base and rotate credentials it could access.
Two malicious versions of Microsoft’s Python package agent-framework briefly appeared on PyPI on November 17, carrying a credential stealer alongside the working framework, according to security advisories. PyPI quarantined both releases, but developers and organizations that installed version 1.26.1 or 1.26.2 should treat the affected environment as exposed and rotate credentials it could reach.
The affected releases were 1.26.1 and 1.26.2. The GitHub Advisory Database says both contained a credential stealer absent from the previous clean release, 1.26.0, and rates the incident critical. The package is the one Microsoft’s installation instructions point to, according to the source report; it is an umbrella package that installs the framework and related components.
According to the timeline reported by StepSecurity, version 1.26.1 was published at 20:35 UTC and quarantined at 21:06 UTC. The attacker uploaded version 1.26.2 at 21:08 UTC, just over two minutes after the first release was quarantined. PyPI quarantined that version at 21:16 UTC. Both releases have been removed from the registry; the source report said the available version was then 1.26.0, published November 12.
The advisory says a startup hook caused the malicious code to run when Python started in an affected environment, even if the user did not import the framework. Researchers described collection of credentials and tokens from developer tools, environment variables and cloud services, as well as requests for certain cloud and pipeline identity tokens. The framework continued to work, which could make the compromise difficult to spot through ordinary use. Security firms including Socket, StepSecurity, Wiz, JFrog and Endor Labs analyzed the 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
Why Build Credentials Were at Risk
The risk extends beyond a developer’s laptop because Python packages are commonly installed in continuous-integration jobs, build servers and container images. Those environments may have access to source repositories, cloud accounts, deployment systems or AI service credentials. If a token was accessible to the affected process, it could have been exposed during the brief publication window.
The malware’s placement in a package that still functioned matters: developers might not see a crash or other obvious sign after installation. The GitHub advisory says the code ran on Python startup without requiring an import of the framework. That means ordinary later activity in the environment, such as running a build or test, could trigger it.
Researchers also highlighted a publishing-control issue. StepSecurity and JFrog said quarantining the first release did not itself revoke the publishing access used to upload it. The follow-up release, posted minutes later, shows why removing a malicious version and securing the account or credential behind its publication are separate response steps.
Products Worth Considering

Logitech MX Brio Ultra HD 4K Webcam

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI Connectivity
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 in Python and coordinating them in workflows. The PyPI package agent-framework acts as an umbrella for the core package and integrations. PyPI is a major public registry used to distribute Python software, so a malicious release there could reach users through direct installation or automated dependency updates.
The clean version immediately preceding the incident was 1.26.0, published November 12, according to the source report. Endor Labs noted that releases it 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 is part of the researchers’ account, not by itself proof of how publishing access was obtained.
The OpenSSF malware database listed the case as MAL-2026-20101, and the GitHub Advisory Database published advisory GHSA-764m-3ppc-5m73. Wiz said it promptly reported its findings to Microsoft on November 17. The source report said Microsoft had given The Register a short statement that it had made some repositories private while investigating malicious content; it found no Microsoft security advisory at the time of publication.
“The credential stealer is not present in the last clean release, 1.26.0.”
— GitHub Advisory Database
Victim Numbers and Access Path
No victim count or confirmed tally of stolen credentials was provided in the source material. It is not clear how many users installed either version, whether any credentials were used after collection, or whether any organization suffered follow-on access.
The source report also does not establish how the attacker obtained publishing access, how many accounts or environments were affected, or whether Microsoft’s investigation identified a broader compromise. StepSecurity reported no remote-code-loading second stage in the code it inspected, and the source report said researchers described the malware as not persistent beyond the installed package. Those findings do not establish that every installation was detected or that every potentially exposed credential is known.
Rebuild, Rotate and Investigate
Organizations should check package lockfiles, build logs, caches and deployment records for installations of 1.26.1 or 1.26.2. The GitHub advisory recommends treating an environment that installed either version as exposed, rebuilding it from a clean base using 1.26.0, and rotating every credential the environment could access, including CI and GitHub Actions secrets.
Security teams should also review relevant account and cloud logs for unusual use of credentials, while developers should avoid relying on package removal alone as a cleanup measure. Microsoft’s investigation and any further advisories could clarify the publishing compromise, impact and recovery guidance; the source report did not give a timetable for further updates.
Key Questions
Which package versions were malicious?
The affected PyPI releases were agent-framework 1.26.1 and 1.26.2. The prior clean release identified by the GitHub Advisory Database was 1.26.0.
Could the malware run if I did not import the framework?
Yes. The advisory says a startup hook ran when Python started in an affected environment, without importing agent_framework.
What should an organization do if it installed an affected release?
Treat the environment as exposed, rebuild it from a clean base on version 1.26.0, and rotate credentials it could access, including CI and GitHub Actions secrets, following the advisory’s guidance.
How many users were affected?
The source material does not provide a confirmed victim count or a tally of credentials stolen. The number of installations and any resulting misuse remain unclear.
Has Microsoft explained how the package was compromised?
Not in the statement described by the source report. Microsoft told The Register it had made some repositories private while investigating malicious content; the method used to gain publishing access was not established.
Source: ThorstenMeyerAI.com
