Understanding The Microsoft AI Agent Framework PyPI Hijack And Response
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Understanding The Microsoft AI Agent Framework PyPI Hijack And Response on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

Two malicious releases of Microsoft’s official Agent Framework Python package, versions 1.26.1 and 1.26.2, appeared on PyPI on November 17, 2026, and were later quarantined. Security advisories say the releases included a credential stealer that could run when Python started; anyone who installed either version should treat the environment as exposed, rebuild it and rotate credentials it could access.

Two malicious releases of Microsoft’s official Python package for its Agent Framework were published to PyPI on November 17, exposing environments that installed them to credential-stealing code, according to security advisories. PyPI quarantined both versions, but the published guidance says affected developers and organizations should rebuild those environments and rotate credentials the code could reach.

The affected releases were agent-framework 1.26.1 and 1.26.2. The GitHub Advisory Database says both contained a credential stealer absent from the last clean version, 1.26.0, and rates the incident critical. The package is the one linked from Microsoft’s installation instructions, rather than an unrelated package using a similar name. Wiz says it receives almost 200,000 downloads a month; that figure is the firm’s reported monthly download count, not a count of affected installations.

According to the timelines published by researchers, version 1.26.1 appeared at 20:35 UTC. PyPI quarantined it at 21:06 UTC, and version 1.26.2 was uploaded at 21:08 UTC. StepSecurity says PyPI quarantined the second release eight minutes later, at 21:16 UTC. The registry’s current version, according to the source article, was 1.26.0, released November 12. Both malicious versions have been removed from availability on PyPI.

The advisory says the code ran through a Python start-up hook, so it could execute when Python started in an affected environment without the framework being imported. Researchers said it sought credentials and tokens from developer and cloud environments. The advisory recommends treating any environment that installed either release as exposed, rebuilding from a clean base using version 1.26.0, and rotating accessible secrets, including CI and GitHub Actions credentials.

At a glance
updateWhen: Published November 17, 2026; both relea…
The developmentPyPI quarantined two malicious releases of Microsoft’s official Agent Framework package after security researchers identified credential-stealing code in 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

Risks to Developer Credentials

The incident matters because Python packages can run inside laptops, automated build systems, CI pipelines and containers that hold cloud keys, source-control tokens and AI service credentials. A compromised development environment may therefore expose access beyond the machine itself. The advisory’s concern is not limited to credentials typed directly into the affected package: it describes collection of secrets available to the environment and requests for certain cloud and pipeline identity tokens.

The package continued to function, according to the source account, so ordinary use might not have produced an obvious failure. That makes installation history and environment logs more useful for determining exposure than relying on visible symptoms. Wiz’s monthly download estimate does not establish how many people installed the two malicious versions, and no victim count is provided in the source material.

The sequence also highlights a registry response issue. Quarantining a release did not immediately stop another upload, researchers said. StepSecurity and JFrog attributed that gap to publishing access not being revoked when the first version was quarantined. That observation is their assessment of the incident; the source material does not provide a complete account of the account or credential compromise behind the uploads.

Ninja Mega Kitchen System Blender and Food Processor, 1500W, 72 oz Pitcher, 8-Cup Bowl, Black (BL770)

Ninja Mega Kitchen System Blender and Food Processor, 1500W, 72 oz Pitcher, 8-Cup Bowl, Black (BL770)

De'Longhi Magnifica Start Automatic Espresso Machine with Milk Frother, Silver

De'Longhi Magnifica Start Automatic Espresso Machine with Milk Frother, Silver

As an affiliate, we earn on qualifying purchases.

How the Two Releases Appeared

Microsoft’s Agent Framework is an open-source Python toolkit for building AI agents and coordinating them in workflows. The PyPI package agent-framework acts as an umbrella package, pulling in framework code and integrations. The source article reports that version 1.26.0 was released November 12, with the two suspect versions uploaded five days later.

The first suspect release was published as a wheel, a ready-to-install package file. Endor Labs said the earlier releases it examined, from 1.21.0 through 1.26.0, had also included source code. After the first version was quarantined, a second upload followed within minutes, with renamed files and additional capabilities, according to StepSecurity. The source account says the second version added GitHub Actions identity-token collection.

Several security firms, including Socket, StepSecurity, Wiz, JFrog and Endor Labs, analyzed the releases. The OpenSSF-run Open Source Malware database listed the case as MAL-2026-20101, and the GitHub Advisory Database published advisory GHSA-764m-3ppc-5m73. These advisories and research reports are the basis for the technical and timeline details summarized here.

“A credential stealer that is not present in the last clean release, 1.26.0”

— GitHub Advisory Database

Victim Count and Investigation Gaps

The source material gives no confirmed number of affected installations or victims. A monthly download estimate for the package cannot show how many users installed versions 1.26.1 or 1.26.2 during their brief availability, or whether an installation led to credential theft. It also does not identify specific organizations or confirmed cases of misuse.

Microsoft’s reported statement says some repositories were made private while it investigated malicious content. The material does not explain which repositories were affected, how the publishing access was obtained, whether credentials were used after collection, or whether any stolen data was accessed. Researchers reported no remote-code-loading second stage in the code they inspected; that finding does not establish whether credentials were exfiltrated or later abused.

The source article reports that the malicious code was not persistent beyond the installed package. However, removing or replacing the package alone would not address credentials it may already have collected. The exact scope of exposure for any individual environment depends on what the package could access while it was installed.

Actions for Affected Environments

Organizations should check package and build records for installations of agent-framework 1.26.1 or 1.26.2, including automatic updates and container builds. The advisory’s stated response is to treat an environment that installed either release as exposed, rebuild it from a clean base on 1.26.0, and rotate every credential it could reach. Teams should include CI and GitHub Actions secrets in that review, as well as relevant cloud, AI-service and package-registry credentials.

Security teams can use the detailed advisories and researcher reports for technical investigation; the source article deliberately omits indicators of compromise and attack details. Microsoft said it was investigating, but the source material reports no full Microsoft security advisory or later findings. Readers should look for further statements from Microsoft, PyPI and the researchers for updates on the publishing access, the number of affected environments and any confirmed downstream misuse.

Key Questions

Which Microsoft Agent Framework package versions were affected?

The affected PyPI releases were agent-framework 1.26.1 and 1.26.2. The GitHub Advisory Database identifies 1.26.0 as the last clean release in the account provided.

Could the malicious code run if the framework was not imported?

Yes. The advisory says a Python start-up hook could run when Python started in an affected environment, without importing agent_framework.

What should someone do if an environment installed either release?

The advisory recommends treating the environment as exposed, rebuilding from a clean base using version 1.26.0, and rotating credentials the environment could access, including CI and GitHub Actions secrets.

How many users were affected?

No confirmed victim or installation count is provided in the source material. Wiz’s estimate of almost 200,000 downloads a month refers to the package’s reported monthly downloads, not installations of the two malicious versions.

Has Microsoft explained how the releases were published?

Not in the reported statement. Microsoft told The Register it had made some repositories private while investigating malicious content. The source material does not give the cause of the publishing-access compromise or the investigation’s findings.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

October 2026: What an Anthropic IPO Actually Unlocks

Anthropic’s planned October 2026 IPO at a valuation of up to $900B marks a major shift in AI industry dynamics, with implications for competition, valuation, and market timing.

/C O R R E C T I O N – DigitalXForce Corporation/

DigitalXForce corrected its Oct. 21 release, naming Arsenal Growth Equity as investor and detailing IDC recognition and reported 514% growth.

What AI Could Mean For Supermarket Freezer Security

More than 115 stores in at least six countries reported refrigeration disruptions overnight. Investigators have not identified the attacker or any AI role.

2026-10-29 – Petra Tschudin / Thomas Moser: Liquidity Against Collateral: The SNB’s Approach

Petra Tschudin and Thomas Moser outlined how SNB collateral rules support liquidity provision, monetary policy and payments at a Geneva event.