🔍 Read the full analysis: Who Should Act After Microsoft’s AI Agent Framework Was Hijacked On PyPI? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Two malicious versions of Microsoft’s Python package agent-framework were published to PyPI on Nov. 17, 2026, and were later quarantined. Advisories say the releases included credential-stealing code; anyone who installed either version should treat the environment as exposed, rebuild from a clean base and rotate credentials it could access.
Two malicious releases of Microsoft’s agent-framework Python package were published on PyPI on Nov. 17, and security advisories say they contained a credential stealer that ran when Python started in an affected environment. PyPI quarantined both releases, but anyone who installed version 1.26.1 or 1.26.2 should treat that environment as exposed and rotate credentials it could reach, according to the GitHub Advisory Database.
The affected releases were 1.26.1, published at 20:35 UTC, and 1.26.2, published at 21:08 UTC. StepSecurity’s timeline says PyPI quarantined version 1.26.1 at 21:06 UTC and version 1.26.2 at 21:16 UTC. The clean release identified by the advisory is 1.26.0, published on Nov. 12; that is the current PyPI version described in the source material.
The GitHub Advisory Database says the malicious code was not present in 1.26.0 and could run without the user importing the framework. The package could still work normally, meaning a successful install or ordinary use would not necessarily alert users that anything was wrong. The advisory rates the incident critical.
The guidance cited in the advisory is to treat systems that installed either affected version as exposed, rebuild them from a clean base pinned to 1.26.0, and rotate every credential they could access. That includes credentials on developer machines, build servers, CI pipelines and containers, as well as CI and GitHub Actions secrets. The affected package is the official package referenced by Microsoft’s installation instructions, not an unrelated lookalike, according to the source account.
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 Secrets Are at Risk
The incident matters beyond users of an AI framework because Python packages are often installed in environments that hold cloud credentials, source-control tokens and deployment secrets. A compromised developer laptop or automated build job may have access to accounts and services beyond the package itself. If credentials were collected, removing the package alone would not invalidate them; teams must assess and rotate exposed secrets.
The reported behavior also complicates detection. The framework reportedly continued to function, while the malicious hook ran when Python started. A developer might therefore see normal results from a build or test and have no immediate sign of infection. Security teams need to check package installation records and affected environments, rather than relying only on visible application failures.
The episode also highlights a weakness in response to compromised publishing access: taking down one release does not, by itself, stop another upload if the publishing credential remains usable. StepSecurity says the second version appeared shortly after the first was quarantined. That sequence makes publisher access control and revocation part of the response, alongside removing malicious files from a registry.
Products Worth Considering

Anker SOLIX C300 Portable Power Station, 288Wh

Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400
As an affiliate, we earn on qualifying purchases.
How the Two Releases Appeared
Microsoft’s Agent Framework is an open-source toolkit for building and coordinating AI agents in Python. The PyPI package agent-framework is an umbrella package that installs the framework’s core code and integrations. The source material says the package is published under Microsoft’s name and is the one linked from Microsoft installation instructions; Wiz estimates it receives almost 200,000 downloads a month. That figure is Wiz’s stated estimate, not a count of confirmed affected installations.
According to the timeline assembled from advisories and researchers, 1.26.1 was uploaded as a wheel-only release, unlike the examined earlier releases that also included source code, according to Endor Labs. After PyPI quarantined it, version 1.26.2 was uploaded with renamed files and additional capabilities, StepSecurity reported. Both versions were subsequently quarantined. The OpenSSF malware database and GitHub Advisory Database then listed the incident, while security companies including Socket, StepSecurity, Wiz, JFrog and Endor Labs analyzed the releases.
The available source account says StepSecurity and JFrog both warned that quarantining the first release did not automatically revoke the compromised publishing credential or prevent a new upload. Wiz said it promptly reported its findings to Microsoft on Nov. 17. Microsoft’s public comment, reported by The Register, was brief: “We have made some repositories private while we investigate malicious content.” The source account reported no Microsoft security advisory at the time it was written.
“The credential stealer is not present in the last clean release, 1.26.0.”
— GitHub Advisory Database
What Investigators Have Not Reported
The number of installations during the exposure window and the number of confirmed victims are not established in the source material. Wiz’s estimate of monthly downloads does not show how many people installed either malicious version, nor whether any credentials were successfully used. No victim count or confirmed follow-on account misuse is provided.
The source account also does not establish who controlled the publishing credential, how it was compromised, or whether all potentially exposed credentials were accessed. Researchers reported findings about the code they inspected, but those observations do not prove that every installation or system behaved identically. Microsoft’s investigation was still underway in the cited statement, and no detailed company advisory was included in the material.
Actions for Maintainers and Users
Users and organizations should first determine whether package records, lockfiles, build logs or container histories show an installation of 1.26.1 or 1.26.2. If either appears, the cited advisory calls for rebuilding from a clean base on 1.26.0 and rotating all credentials accessible from the affected environment. Teams should include cloud, source-control, package-registry and CI credentials in that review, and check relevant accounts for unexpected activity.
Microsoft’s next public update may clarify how the publishing access was obtained, what repositories were made private and whether any misuse of credentials has been confirmed. The source material does not give a timetable for that update. Package maintainers and registry operators will also need to establish whether publishing credentials were revoked and what safeguards will prevent further releases. Until those details are available, the clearest immediate step for users is to verify installed versions and treat affected environments as exposed.
Key Questions
Which versions of Microsoft’s package were affected?
The GitHub Advisory Database identifies agent-framework 1.26.1 and 1.26.2 as malicious. It identifies 1.26.0, released Nov. 12, as the last clean version.
Did users have to import the framework for the malicious code to run?
No. The advisory says the code could run when Python started in an affected environment, without importing agent_framework. The framework could continue to work normally.
What should an organization do if it installed an affected release?
The cited advisory recommends treating the environment as exposed, rebuilding from a clean base using version 1.26.0, and rotating every credential the environment could access, including CI and GitHub Actions secrets.
How many people were affected?
The source material does not provide a confirmed victim or installation count. Wiz’s estimate of almost 200,000 monthly downloads is not a count of installs of the two malicious releases.
Has Microsoft explained how the package was compromised?
Not in the material provided. Microsoft said it had made some repositories private while investigating malicious content, but the source account reported no detailed public advisory or explanation at that time.
Source: ThorstenMeyerAI.com
