AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Versions 1.26.1 and 1.26.2 of Microsoft’s official agent-framework Python package contained credential-stealing code, according to security advisories. PyPI quarantined both releases, but anyone who installed either should treat reachable credentials as exposed, rebuild from a clean version and rotate secrets.

Two releases of Microsoft’s official Agent Framework Python package contained credential-stealing code, according to the GitHub Advisory Database and security researchers who analyzed the packages. The versions, 1.26.1 and 1.26.2, were published to PyPI on November 17 and quarantined that evening; developers and organizations that installed either should treat credentials accessible to those environments as exposed.

Before you orderOffer from Amazon

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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

The advisory identifies 1.26.1 and 1.26.2 as malicious and says the credential stealer was absent from the last clean release, version 1.26.0. The first release appeared at 20:35 UTC; the second was published at 21:08 UTC. StepSecurity reports that PyPI quarantined the first at 21:06 and the second at 21:16. Both versions have since been removed from PyPI, where 1.26.0 was listed as the current release when the project data was checked on November 19.

The malicious packages included a startup hook, according to the advisory. It ran when Python started in an affected environment, even if the user did not import the Agent Framework. The framework continued to work, which could make the compromise difficult to notice through ordinary application behavior. The advisory says the code sought credentials, including cloud keys, CI tokens and AI service credentials.

Security firms including Socket, StepSecurity, Wiz, JFrog and Endor Labs analyzed the releases. The GitHub Advisory Database rates the incident critical. Wiz says it reported its findings to Microsoft on November 17. Microsoft told The Register: “We have made some repositories private while we investigate malicious content.”

At a glance
reportWhen: Published November 17, 2026; both relea…
The developmentTwo malicious versions of Microsoft’s agent-framework package were published on PyPI on November 17 and were later quarantined.
Bar chart titled 50 stored requests: who sent them? Based on Wiz's analysis of the requests stored by the attacker's free collection service: 29 from sandboxes, vendor scans and tests; 17 carrying real victim data; 4 from the attacker's own machines. Highlighted: behind the 17 requests, 3 victim cases, which Wiz counts as unique, successful thefts of victim data. Note: only what the free service stored, at most 50 requests; no install count has been published
What the attacker's collection service stored, per Wiz Research (researchers' claim, updated November 18, 2026). Chart: Thorsten Meyer AI.

Credentials Reach Beyond One Package

The risk extends to any credentials an affected laptop, build server, CI pipeline or container could access. A compromised token can grant access to connected services or repositories, depending on its permissions. The advisory specifically calls out CI and GitHub Actions secrets among the credentials users should rotate.

Because the code ran at Python startup and the framework remained functional, a working application does not establish that its environment was clean. Nor does PyPI’s removal of the releases remove copies already present in local caches, mirrors or container images. Teams need to check where the versions were installed and what credentials those environments could reach.

The timing of the second upload also matters for package-registry security. StepSecurity says it appeared two minutes and 23 seconds after the first release was quarantined. Researchers say quarantining a version did not immediately revoke the publishing access used to upload it. The incident therefore raises questions about both the initial compromise and how quickly publishing credentials were disabled.

Products Worth Considering

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

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

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

As an affiliate, we earn on qualifying purchases.

A Two-Release PyPI Timeline

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 that brings in the framework’s core code and integrations. Microsoft’s installation instructions point users to this package, making the releases part of the normal software supply chain for developers using the toolkit.

The last clean version identified in the advisory was 1.26.0, published November 12. On November 17, version 1.26.1 appeared as a wheel, a ready-to-install Python package file. StepSecurity says PyPI quarantined it at 21:06 UTC, about 31 minutes after publication. Version 1.26.2 followed at 21:08 and was quarantined at 21:16, according to the same firm. The advisory says the second release used renamed files and also targeted GitHub Actions tokens.

The OpenSSF malware database listed the incident as MAL-2026-20101 on November 17. The GitHub Advisory Database published GHSA-764m-3ppc-5m73 on November 18. The listed sources describe two poisoned versions and identify 1.26.0 as the clean version at the time of their reports.

““We have made some repositories private while we investigate malicious content.””

— Microsoft, in a statement to The Register

Exposure Scope Remains Unknown

The available material does not state how many people or organizations installed either release, or how many credentials were accessed or misused. Wiz says the package receives almost 200,000 downloads a month, but that figure is not an estimate of downloads of the malicious versions, and the sources do not provide a count for the incident window.

Microsoft’s statement does not identify the repositories it made private, explain how publishing access was obtained, or say whether its investigation has determined the incident’s full scope. The advisories and researcher reports describe the package behavior, but the supplied information does not confirm that every credential reachable from every affected environment was taken or used.

PyPI’s quarantine prevents installation through the registry while the releases remain quarantined. It does not establish whether the versions persist in private mirrors, cached package stores, build artifacts or deployed images. Organizations will need to check their own installation records and infrastructure.

Rebuild and Rotate Exposed Secrets

Users should check package records, lockfiles, build logs and container histories for agent-framework 1.26.1 or 1.26.2. The advisory recommends treating an environment that installed either version as exposed, rebuilding it from a clean base using version 1.26.0, and rotating every credential the environment could reach. That includes CI and GitHub Actions secrets, as well as cloud and AI service credentials within its access.

Teams should also identify cached or mirrored copies and replace images or artifacts built from an affected environment. Rotating credentials without removing compromised environments may leave the malicious code in place; rebuilding without rotating reachable secrets may leave credentials exposed. The recommended response addresses both the environment and the access it held.

Microsoft’s investigation and any further updates from the company or security researchers may clarify how the publishing access was compromised and whether there is evidence of credential misuse. The supplied sources do not give a date for a further advisory or an expected investigation milestone. Until more information is published, affected organizations must determine their own exposure from package and build records.

Key Questions

Which versions of Microsoft’s Agent Framework package were affected?

The GitHub Advisory Database identifies agent-framework 1.26.1 and 1.26.2 as malicious. It identifies 1.26.0 as the last clean release.

What should an organization do if it installed one of the affected versions?

Treat the environment as exposed, rebuild from a clean base using version 1.26.0, and rotate credentials that the environment could access. The advisory specifically includes CI and GitHub Actions secrets.

Did PyPI remove the malicious releases?

PyPI quarantined both releases on November 17, according to StepSecurity’s timeline. Quarantine does not remove copies that may already exist in caches, mirrors, build artifacts or container images.

How many users were affected?

The supplied advisories and researcher reports do not state how many users installed the malicious releases or how many credentials were accessed. Wiz’s figure of almost 200,000 monthly downloads is not a count of downloads of versions 1.26.1 and 1.26.2.

Has Microsoft explained how the package was compromised?

Microsoft told The Register that it had made some repositories private while investigating malicious content. The supplied material does not explain how publishing access was obtained or provide the investigation’s conclusions.

Source: Thorsten Meyer AI

You May Also Like

How Software Engineering Is Changing: A Tale Of Two Eras

Neil Kakkar says he merged 11 times more pull requests at AI-native startup Tano than at PostHog, alongside new costs in focus and coordination.

I Tried Google’s New Image Editor, And It Could Replace Canva And Photoshop

Google’s new prompt-driven image editor Google Pics blends Canva-style design and AI photo editing for AI Pro and Ultra subscribers.

4 Best Content Creator Laptops In 2027

A comparison of four Windows laptops for creators weighs memory, processors, screen size and graphics limits using the supplied product details.

I Tried All The Best (And Worst) Doorbells

A comprehensive review of popular and unpopular doorbells, exploring their features, performance, and user experiences amid rising interest in home security tech.