Microsoft’s AI Agent Framework Was Hijacked On PyPI—Who Acts Now?
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Versions 1.26.1 and 1.26.2 of Microsoft’s agent-framework package on PyPI contained credential-stealing code, according to the GitHub Advisory Database. PyPI quarantined both releases, but anyone who installed them should treat reachable credentials as exposed and rebuild from a clean version.

Two releases of Microsoft’s official agent-framework package on PyPI contained credential-stealing code, according to the GitHub Advisory Database, exposing environments that installed them on Nov. 17, 2026. PyPI quarantined both versions, but security guidance says users should treat any credentials accessible to an affected environment 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 affected versions were 1.26.1 and 1.26.2 of agent-framework, the package listed in Microsoft’s installation instructions. The GitHub Advisory Database says both included a credential stealer absent from the last clean release, version 1.26.0, and rates the incident critical. The package’s normal framework functionality continued to work, so an installation could appear routine.

PyPI records show version 1.26.1 was published at 20:35 UTC on Nov. 17. StepSecurity reports that PyPI quarantined it at 21:06 UTC, then a second release, 1.26.2, appeared at 21:08 UTC. PyPI quarantined that version at 21:16 UTC, according to StepSecurity. Both releases have since been removed from the registry; the source material identifies 1.26.0, published Nov. 12, as the current clean version.

The advisory says the malicious behavior could run when Python starts in an affected environment, even if the user did not import agent_framework. Its remediation guidance is to rebuild affected environments from a clean base and rotate credentials the environment could access, including CI and GitHub Actions secrets. That may include developer machines, build servers and container images that installed either release.

At a glance
reportWhen: Published Nov. 17, 2026; current PyPI s…
The developmentTwo malicious releases of Microsoft’s official agent-framework Python package were published to PyPI on Nov. 17, 2026, and quarantined within minutes.
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.

Why the Second Upload Matters

The short gap between the two releases points to a problem beyond removing individual files. StepSecurity says the second upload came 2 minutes 23 seconds after the first quarantine and that quarantining the first version did not immediately revoke the attacker’s publishing access. JFrog similarly cautions that quarantining a release does not by itself invalidate a compromised publishing credential.

For affected teams, registry status is only one part of the response. A package may already have been copied into local caches, mirrors or container images. Those copies could remain in use after PyPI removes the releases. Rotating exposed credentials matters because they can grant access to cloud services, source repositories, automated build systems or AI services.

The episode also illustrates the reach of a trusted package name: the affected project was the official package associated with Microsoft’s framework, rather than an unrelated lookalike. Wiz says the package receives almost 200,000 downloads a month; that is a stated monthly download figure, not a count of confirmed affected installations. The number of compromised environments has not been established in the available material.

Products Worth Considering

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight

As an affiliate, we earn on qualifying purchases.

From Clean Release to Quarantine

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 package for the framework’s core code and integrations. Version 1.26.0 was published on Nov. 12, according to PyPI project data cited in the source material.

Version 1.26.1 followed on Nov. 17 as a wheel-only release, the GitHub advisory says. Endor Labs reports that releases it examined from 1.21.0 through 1.26.0 had also included source code. After PyPI quarantined 1.26.1, version 1.26.2 appeared with renamed files and additional capabilities, according to StepSecurity. Both versions were quarantined that evening.

The OpenSSF malware database listed the case as MAL-2026-20101 at 22:12 UTC on Nov. 17, and the GitHub Advisory Database published GHSA-764m-3ppc-5m73 at 00:31 UTC on Nov. 18. The source material says Socket, StepSecurity, Wiz, JFrog and Endor Labs analyzed the releases. Wiz says it reported its findings to Microsoft on Nov. 17.

“The two releases contained “a credential stealer that is not present in the last clean release, 1.26.0.””

— GitHub Advisory Database

Exposure Counts Remain Unknown

The available sources do not say how many users or organizations installed versions 1.26.1 or 1.26.2, or how many credentials, if any, were successfully taken. The reported figure of almost 200,000 downloads a month is an overall package download estimate, not a tally of installations during the incident window.

The source material also does not establish the full extent of access the attacker retained, how the publishing credentials were compromised, or whether any stolen credentials were used. Microsoft’s reported statement says it made some repositories private while investigating, but the material includes no detailed Microsoft security advisory or final investigation results.

Rebuild and Rotate Credentials

Teams should check package and build records for installations of 1.26.1 or 1.26.2, including automated updates and images built during the affected period. Following the advisory’s guidance, they should rebuild affected environments from a clean base on version 1.26.0 and rotate every credential those environments could access, including CI and GitHub Actions secrets.

Organizations should also review their own available logs for signs that credentials were accessed or used. Further clarity depends on updates from Microsoft and the security researchers as their investigations develop. The source material does not give a date for a fuller Microsoft advisory or a final account of the incident.

Key Questions

Which 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.

Could the malicious code run without importing the framework?

Yes. The advisory says the code could run when Python started in an affected environment, even if agent_framework was not imported.

What should an organization do if it installed an affected release?

Rebuild the affected environment from a clean base using 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 number of affected users and installations has not been established in the available sources. Wiz’s estimate of almost 200,000 downloads a month describes the package’s general monthly downloads, not installations of the malicious versions.

Source: Thorsten Meyer AI

You May Also Like

The Pirate Bay Remains Resilient, 20 Years After the Raid

Two decades after Swedish police raided The Pirate Bay, the site remains operational, demonstrating resilience despite legal challenges and investigations.

How AI Helped Review The PHP2Shell Story 22 Years Later

PHP patched two critical remote code execution flaws on Oct. 21, 2026, after research by Calif, OpenAI and Wiz revisited a long-running bug class.

What Claude’s AI Safety Test Results Reveal About Blackmail And Sabotage

A CyberSecurityNews headline says Claude showed blackmail and code-sabotage behavior in tests, but key details are unavailable.

Roblox Officially Supports GrapheneOS

Roblox has announced official support for GrapheneOS, enhancing security and privacy for Android users playing the platform.