The Microsoft AI Agent Framework PyPI Incident: Who Must Respond?
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 package on PyPI contained credential-stealing code, according to the GitHub Advisory Database. 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 package on PyPI contained code that stole credentials, according to the GitHub Advisory Database and security researchers. The releases, versions 1.26.1 and 1.26.2, were uploaded on November 17 and quarantined after they were detected; anyone who installed either may have exposed credentials available to that environment.

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

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

The GitHub Advisory Database says both releases contained a credential stealer absent from clean version 1.26.0 and rates the incident critical. The code ran when Python started in an affected environment, even if the user did not import the Agent Framework package. The framework itself continued to work, which could make the compromise less visible to users.

PyPI records show version 1.26.1 appeared at 20:35 UTC on November 17. StepSecurity says PyPI quarantined it at 21:06 UTC, about 31 minutes later. The attacker then uploaded version 1.26.2 at 21:08 UTC; PyPI quarantined that release at 21:16, according to StepSecurity. The second upload included renamed files and additional capabilities, researchers reported.

The advisory recommends treating environments that installed either version as exposed, rebuilding from a clean base using 1.26.0, and rotating every credential those environments could access, including CI and GitHub Actions secrets. Microsoft told The Register: “We have made some repositories private while we investigate malicious content.”

At a glance
reportWhen: Incident occurred November 17, 2026; ad…
The developmentTwo malicious releases of Microsoft’s official Agent Framework Python package were uploaded to PyPI on November 17 and 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.

Why the Two-Minute Reupload Matters

The incident affected the official package linked from Microsoft’s installation instructions, rather than an unrelated package using a similar name. Wiz says the project receives almost 200,000 downloads a month; that figure is the firm’s reported monthly download count, not an estimate of installations during the incident window.

The brief gap between quarantine and the second upload points to a key response issue: removing a release does not necessarily revoke the publishing credential used to upload it. StepSecurity and JFrog said the attacker was able to publish again after the first version was quarantined. For users, quarantine also cannot remove copies already present in local caches, mirrors, containers or build artifacts.

Credential exposure can extend beyond a developer’s workstation. Build systems and CI pipelines may hold cloud keys, repository tokens and AI service credentials. Rotating credentials and replacing affected environments can limit further access, but the source material does not establish whether any stolen credentials were used after collection.

Products Worth Considering

Kindle Paperwhite Signature Edition 32GB (2024 Model) - Metallic Black

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

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

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

Apple MacBook Air 13-inch with M5 Chip, 16GB Memory, 512GB SSD, Liquid Retina Display, and 12MP Center Stage Camera – Midnight

Apple MacBook Air 13-inch with M5 Chip, 16GB Memory, 512GB SSD, Liquid Retina Display, and 12MP Center Stage Camera – Midnight

As an affiliate, we earn on qualifying purchases.

How the Releases Reached PyPI

Microsoft’s Agent Framework is an open-source toolkit for building AI agents and coordinating workflows in Python. The PyPI package agent-framework is an umbrella package that brings in framework components and integrations. Its legitimate version 1.26.0 was published on November 12 and remained the current PyPI version when project data was checked on November 19.

According to the GitHub advisory, version 1.26.1 was uploaded as a wheel, a ready-to-install package file. Endor Labs said releases it examined from 1.21.0 through 1.26.0 had also included source code. Researchers from Socket, StepSecurity, Wiz, JFrog and Endor Labs analyzed the malicious releases. The OpenSSF malware database listed the case as MAL-2026-20101 on November 17; the GitHub Advisory Database published GHSA-764m-3ppc-5m73 early on November 18.

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

— GitHub Advisory Database

What Investigators Have Not Confirmed

The available source material does not establish how the attacker obtained publishing access, how many users installed either release, or whether stolen credentials were used. It also does not provide a confirmed count of affected organizations or systems. Monthly download figures do not show how many installations occurred during the releases’ exposure windows.

Microsoft’s statement describes an investigation but does not give a detailed incident report in the material provided. It is not clear whether the publishing access has been fully revoked, what repositories were made private, or whether additional affected releases or systems have been identified.

What Users and Microsoft Must Do

Users and organizations should identify whether their laptops, build servers, CI pipelines or container images installed agent-framework 1.26.1 or 1.26.2. The advisory’s guidance is to rebuild affected environments from a clean base on 1.26.0 and rotate every credential those environments could reach, including CI and GitHub Actions secrets. Teams should also check cached packages and images rather than relying only on PyPI’s current listing.

Microsoft’s investigation and further advisories may clarify how the publishing access was compromised, whether credentials were misused and what steps were taken to prevent further uploads. Until those details are published, the number of affected installations and the full scope of any credential exposure remain unknown.

Key Questions

Which versions of the 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 in this incident.

Does using the framework require importing it for the malware to run?

No. The advisory says the added code ran when Python started in an affected environment, without importing agent_framework.

What should an organization do if it installed either release?

Treat the environment as exposed, rebuild it from a clean base using 1.26.0, and rotate all credentials it could access, including CI and GitHub Actions secrets, following the advisory’s guidance.

Were any credentials confirmed stolen or misused?

The source material describes credential-stealing code but does not confirm how many credentials were collected or whether any were used after collection.

Source: Thorsten Meyer AI

You May Also Like

A Startup Replay Lets QA Readers Inspect AI Business Decisions

The AI Company Emulator replays an emulated AI team running the construction-site app GewerkTon day by day, every decision visible and clearly labelled as a simulation.

The Impact Of AI On Evolving Work Environments

OpenAI publishes an article titled ‘How workers are unlocking new ways of working,’ highlighting evolving workplace practices amid AI adoption. Details pending.

The Subsidy Behind The Number: SemiAnalysis On AI Subscriptions

A SemiAnalysis audit finds Claude subscriptions can offer roughly five to six times ChatGPT’s API-equivalent usage on selected models, while limits and costs shift.

ByteDance’s New AI Division Focuses On Data Post-Seed And Flow Success

ByteDance has reportedly created a new AI unit centered on data, expanding its organizational structure after Seed and Flow, though details remain undisclosed.