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

🔍 Read the full analysis: Microsoft’s AI Agent Framework Was Hijacked On PyPI. Who Responds? on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

  • Fast, free delivery for dorm and study essentials
  • Prime Video and Amazon Music included
  • Member-only deals
Try Prime for Young Adults Free trial for eligible 18–24 year olds
As an affiliate, we earn on qualifying purchases.

TL;DR

Two versions of Microsoft’s Python package agent-framework, 1.26.1 and 1.26.2, carried credential-stealing code and were briefly available on PyPI on Nov. 17, 2026. PyPI quarantined both releases, but developers and organizations that installed either should treat reachable credentials as exposed, rebuild affected environments and rotate those credentials.

Two malicious versions of Microsoft’s agent-framework Python package were briefly available on PyPI on November 17, carrying code that security researchers say stole credentials from affected environments. The Python package registry quarantined both releases, but anyone who installed version 1.26.1 or 1.26.2 should treat the machine or build environment as exposed and rotate credentials it could access, according to the GitHub Advisory Database.

The affected package, agent-framework, is the official PyPI package for Microsoft’s open-source toolkit for building AI agents and coordinating workflows. 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 continued to function, according to the source advisory, so users may not have seen an obvious failure after installation.

PyPI records and security researchers place the first upload at 20:35 UTC on November 17. StepSecurity says PyPI quarantined it at 21:06 UTC. A second version appeared at 21:08 UTC and was quarantined at 21:16 UTC. Both releases have since been removed from the registry; the source material identifies 1.26.0, released November 12, as the current version there.

The advisory says the malicious code ran when Python started in an affected environment, even if a user did not import the framework itself. It sought credentials and tokens available to the environment, including cloud, developer-tool and AI-service secrets. StepSecurity reported finding no remote-code-loading second stage in the code it inspected. The source material says several security firms, including Socket, StepSecurity, Wiz, JFrog and Endor Labs, analyzed the releases.

At a glance
reportWhen: Published November 19, 2026; malicious…
The developmentResearchers reported that malicious releases of Microsoft’s agent-framework package were published on PyPI before the registry quarantined them.

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

  1. Clean 1.26.0 publishedStill the current version on PyPI
  2. Malicious 1.26.1 appearsFramework works normally, a credential stealer runs whenever Python starts
  3. PyPI quarantines 1.26.1After about 31 minutes
  4. Attacker uploads 1.26.2Renamed files, also targets GitHub Actions tokens
  5. PyPI quarantines 1.26.2Eight minutes later
  6. Malware databases list itMAL-2026-20101 (OpenSSF), GHSA-764m-3ppc-5m73 (GitHub)
Quarantine is not revocation. The second release came 2 minutes 23 seconds after the first was pulled: the attacker still had publishing access. Copies already in caches, mirrors and images stay there.

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 Are Exposed

The incident matters because developer machines and automated build systems often hold credentials that grant access to source code, cloud infrastructure and deployment pipelines. If a malicious package can read those secrets, a brief installation window may create risk beyond the computer where it ran. The GitHub Advisory Database specifically warns users to rotate credentials the affected environment could reach, including CI and GitHub Actions secrets.

The package’s normal operation also complicates detection. The advisory says the framework kept working while the code ran at Python startup, meaning a successful installation or routine test would not, on its own, show that the environment was safe. Wiz says the package receives almost 200,000 downloads a month; that is a reported monthly download figure, not a count of affected installations or victims.

For organizations, the response is not limited to removing the package. Credentials that may have been accessible need review and rotation, and affected machines, containers and build environments need to be rebuilt from a clean base. Those steps reduce the chance that stolen secrets remain useful, although the source material does not establish whether any were used after collection.

KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)

KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)

Ninja Foodi Smart 6-in-1 Indoor Grill and Air Fryer, 4 qt, FG551

Ninja Foodi Smart 6-in-1 Indoor Grill and Air Fryer, 4 qt, FG551

As an affiliate, we earn on qualifying purchases.

How the Two Releases Appeared

The clean package version 1.26.0 was published on November 12. The source material says the first malicious release, 1.26.1, was uploaded as a wheel only, unlike earlier releases examined by Endor Labs, which also included source code. After PyPI quarantined it, an attacker published 1.26.2 roughly two minutes later. StepSecurity says the second release used renamed files and added capabilities.

That sequence points to a gap between removing a malicious release and stopping further publishing. StepSecurity said quarantining version 1.26.1 did not immediately revoke the publishing access used to upload it. JFrog likewise said removing one version does not, by itself, revoke a compromised publishing credential or prevent another upload. OpenSSF’s malware database listed the case as MAL-2026-20101; the GitHub Advisory Database later published advisory GHSA-764m-3ppc-5m73.

Microsoft’s Agent Framework is an open-source Python toolkit. The PyPI package named agent-framework serves as an umbrella package that pulls in the core framework and integrations. The fact that the malicious versions were on the official package name, rather than an unrelated lookalike, is why users following the project’s standard installation route could have been affected.

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

— Microsoft, in a statement to The Register

Victim Counts Remain Unknown

The source material gives no verified count of installations, affected organizations or credentials taken. Wiz’s estimate of almost 200,000 monthly downloads is not an estimate of downloads during the incident window, and it does not show how many users installed either malicious release. It is also not clear whether any stolen credentials were used or whether data was accessed beyond the collection described by researchers.

The source material reports that Microsoft said it had made some repositories private while investigating, but does not identify them or provide a fuller public incident account. It also says no Microsoft security advisory had been found at the time of publication. The precise path by which the attacker obtained publishing access is not established in the material provided.

Rebuild, Rotate and Monitor

Developers and organizations should check package records, lockfiles, build logs and container histories for installations of 1.26.1 or 1.26.2. The GitHub Advisory Database’s guidance, as reported in the source material, is to treat any environment that installed either version as exposed, rebuild it from a clean base using the clean 1.26.0 release, and rotate every credential that environment could access.

That review should include CI and GitHub Actions secrets, cloud credentials, package-registry access and AI-service keys where present. Teams should also check relevant account and cloud activity for suspicious use of credentials during and after the exposure window. The source material does not give a confirmed victim list or a date for a fuller Microsoft update, so affected users should follow the GitHub advisory, PyPI package records and Microsoft’s subsequent statements for new information.

Key Questions

Which versions of Microsoft’s package were malicious?

The source advisory identifies agent-framework 1.26.1 and 1.26.2 as malicious. It identifies 1.26.0, released November 12, as the last clean version.

What should I do if I installed one of the releases?

Treat the environment as exposed. Rebuild it from a clean base using the clean release, and rotate credentials the environment could reach, including relevant cloud, CI, GitHub Actions, package-registry and AI-service secrets.

Did the package have to be imported for the code to run?

No. The GitHub Advisory Database says the code ran when Python started in an affected environment, even without importing agent-framework.

How many users were affected?

No verified victim or installation count is provided in the source material. Wiz’s reported figure of almost 200,000 downloads a month is not a count of downloads of the two malicious releases.

Has Microsoft explained how the attacker published the releases?

The source material does not establish how publishing access was obtained. It reports that Microsoft told The Register it had made some repositories private while investigating malicious content.

Source: ThorstenMeyerAI.com

You May Also Like

Malware On 178 Websites: What AI Users Can Do To Review Their Exposure

Report URI says malicious code was served through Optimizely to visitors on 178 organisations’ sites on Nov. 16. Here is what is known and what users can do.

Technology Operations Signal Monitor: How Google Helped Destroy Adoption Of RSS Feeds (2023)

New insights show how Google’s platform and tooling changes contributed to the decline of RSS feed adoption, impacting small software companies.

7 Best Office Product Scanners for Prime Day Deals in 2026

Explore the best office product scanners on Prime Day 2026, including top picks for shared and solo offices, with detailed insights on features and value.

Broccoli2Shell: How To Check Your Applications After Applying A Patch

A Brotli decoder flaw affects multiple runtimes and tools. See confirmed fixes, exposure paths and why patched applications must be restarted.