🔍 Read the full analysis: The Microsoft AI Agent Framework PyPI Hijack: Roles And Responsibilities on ThorstenMeyerAI.com
Get audio and creator gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
PyPI versions 1.26.1 and 1.26.2 of Microsoft’s agent-framework package contained credential-stealing code, according to security advisories and researchers. The releases were quarantined on November 17, 2026, but the number of affected installations and any stolen credentials remain unknown. Advisories tell users to rebuild affected environments and rotate credentials those environments could access.
The two releases appeared at 20:35 UTC and 21:08 UTC, respectively. StepSecurity says PyPI quarantined version 1.26.1 at 21:06 UTC, then quarantined 1.26.2 at 21:16 UTC. The source account describes the first release as available for about 31 minutes. PyPI’s current listed version is 1.26.0, released November 12, according to the supplied source.
The GitHub Advisory Database says both malicious releases contained a credential stealer absent from the last clean release, 1.26.0, and rates the incident critical. The code ran through a Python start-up hook, so it could execute when Python started in an affected environment without the user importing the framework itself. The framework continued to work, which could make the compromise less apparent to users.
According to the advisory, the code sought credentials in environment variables and developer-tool files, and could request tokens from cloud and CI services. Version 1.26.2 added GitHub Actions identity-token access, the source reports. Researchers including Socket, StepSecurity, Wiz, JFrog and Endor Labs analyzed the releases. The supplied account says PyPI removed both, but removal does not establish whether any credentials were accessed or misused.
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
Credential Exposure Across Developer Systems
The risk extends beyond a developer’s workstation. Python environments used for build servers, CI pipelines and container images may hold cloud keys, source-control tokens or AI-service credentials with access to production systems and private code. A package can therefore work as intended while its installation creates a separate security exposure.
Security teams should not treat package removal as a complete remedy. The advisory’s recommended response is to identify environments that installed either version, rebuild from a clean base using version 1.26.0, and rotate every credential the affected environment could reach, including CI and GitHub Actions secrets. Whether attackers used any collected credentials is not established in the supplied reporting.
Products Worth Considering

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

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

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
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 workflows. The PyPI package agent-framework is an umbrella package that brings in the core framework and integrations. The source says Microsoft’s installation instructions direct users to this official package, rather than an unrelated lookalike.
After the first release was quarantined, a second appeared about two minutes later. StepSecurity says quarantine did not immediately revoke the attacker’s publishing access; JFrog likewise notes that removing one release does not itself invalidate a compromised publishing credential. The supplied account does not establish how that access was obtained or identify a responsible individual.
The OpenSSF malware database listed the case as MAL-2026-20101 on November 17, and the GitHub Advisory Database published GHSA-764m-3ppc-5m73 early on November 18. Wiz said it reported its findings to Microsoft on November 17. The Register reported Microsoft’s brief response that it had made some repositories private while investigating malicious content. The supplied source says it found no Microsoft security advisory at the time of its publication, November 19.
“The malicious code is not present in the last clean release, 1.26.0.”
— GitHub Advisory Database
Victim Count and Credential Use
The supplied source does not give a confirmed number of downloads or affected environments, nor does it report confirmed credential theft, account access or downstream breaches. Wiz’s estimate of almost 200,000 downloads a month describes the package’s usual download volume, not the number of downloads during the incident or the number of affected users.
It is also unclear how the attacker gained publishing access, whether credentials were collected from any installations, and whether any were used before expiration or rotation. StepSecurity reported no remote-code-loading second stage in the code it inspected, and The Register described the malware as not persistent beyond the installed package; these findings do not establish that affected credentials were safe.
Rebuild, Rotate and Monitor
Organizations should check package records, lockfiles, build logs and container histories for agent-framework 1.26.1 or 1.26.2. If either version was installed, the advisory calls for rebuilding the environment from a clean base on 1.26.0 and rotating credentials it could access. Teams should also review cloud, CI and source-control logs for suspicious token use, while following the detailed security advisories for technical indicators and response steps.
The supplied reporting does not state when Microsoft will publish further findings or whether a separate Microsoft security advisory is planned. Further information about the publishing-access compromise, affected installations and any credential misuse remains pending.
Key Questions
Which package versions were affected?
The agent-framework PyPI versions 1.26.1 and 1.26.2 contained credential-stealing code, according to the GitHub Advisory Database. The source identifies 1.26.0 as the last clean release.
Could the malware run without importing the framework?
Yes. The advisory says a Python start-up hook could run when Python started in an affected environment, without importing agent_framework.
What should users do if they installed an affected version?
Treat the environment as exposed, rebuild it from a clean base using version 1.26.0, and rotate credentials it could access, including relevant cloud, CI and GitHub Actions secrets. Review relevant service logs for suspicious activity.
How many people were affected, and were credentials used?
The supplied reporting gives no confirmed victim count and does not confirm that stolen credentials were used. The package’s reported monthly download estimate is not a count of downloads during the incident or affected installations.
Has Microsoft explained how the releases were published?
Not in the supplied reporting. Microsoft told The Register it had made some repositories private while investigating malicious content, but the source gives no finding about how publishing access was obtained.
Source: ThorstenMeyerAI.com
