Agent Framework PyPI Malware
Attackers published malicious releases of Microsoft's official Agent Framework package on PyPI on November 17, 2026. Version 1.26.1 harvested secrets, with a focus on Azure. This post covers what is known, what is not, and what to check.
Where Worldwide
On this page 9 sections
Attackers published malicious releases of Microsoft's official Agent Framework Python package on PyPI on November 17, 2026. One of them, version 1.26.1, contained a credential stealer that took secrets from environment variables and local configuration files. If anyone who writes code for your company, staff or contractor, installed or updated that package that evening, assume the keys on that machine were stolen and replace them.
What happened to Microsoft's Agent Framework package on PyPI
PyPI is the public registry most Python projects download libraries from. BleepingComputer reported that attackers compromised Microsoft's official Agent Framework package there and pushed malicious releases on November 17, 2026.
Wiz researchers report that the malicious version 1.26.1 went up at 20:35 UTC that day. PyPI quarantined the first malicious release, and Wiz says that as of 21:30 UTC, both malicious versions were no longer available. So 1.26.1 was up for less than an hour.
A short window does not mean nobody was exposed. BleepingComputer says the package gets roughly 200,000 monthly downloads, and OX Security counts 192,534. Those figures describe normal use of the package; they do not say how many people installed the bad versions, and none of our sources gives that number.
What is confirmed and what is still unclear
| Statement | Where it comes from | How solid it is |
|---|---|---|
| Version 1.26.1 contained a credential stealer | BleepingComputer, Wiz | Reported by researchers who examined it |
| Version 1.26.1 was published at 20:35 UTC on November 17, 2026 | Wiz | Researcher timeline |
| Both malicious versions were gone from PyPI by 21:30 UTC | Wiz | Researcher observation |
| The most recent safe version is 1.26.0 | Wiz | Researcher guidance |
| The stealer focuses on Azure and Azure DevOps | Wiz | Researcher analysis |
| The attacker appears to have compromised a client certificate and private key for Microsoft ESRP | Wiz | Wiz's assessment, worded as "appears" |
| How the attacker got publishing access | BleepingComputer | Not known |
| Microsoft's own account of the breach | BleepingComputer has asked Microsoft | No Microsoft statement in the sources we used |
How did attackers publish under Microsoft's name?
Nobody has confirmed this yet. BleepingComputer states: "It remains unclear how the attacker gained access to publish malicious releases." The outlet has contacted Microsoft to find out where the breach happened.
Wiz offers a lead: the attacker appears to have compromised a client certificate and private key for Microsoft ESRP. That is Wiz's assessment, not a confirmed finding, and our sources do not say how the certificate was obtained or what else it may have unlocked.
Wiz refers to two malicious versions and names 1.26.1; the second version number is not in our sources. Wiz says 1.26.0 is the most recent safe version, so treat anything newer from that day as suspect.
What the Agent Framework stealer was after
The stealer collected secrets from environment variables and local configuration files. Developers often keep API keys, cloud tokens and database passwords in exactly those places, so their tools can sign in without asking.
Wiz adds: "The stealer is focused on Azure and Azure DevOps." Any machine that ran 1.26.1 while holding Azure or Azure DevOps credentials is the main concern, and other secrets on that machine should be treated as exposed too.
The malicious tooling identifies itself as Cairn, which Wiz describes as an open-source AI penetration-testing framework. That is what the tool calls itself; it does not tell us who ran the attack, and none of our sources names an attacker.
Who is affected by the Agent Framework 1.26.1 compromise
- Developers who installed or upgraded Microsoft's Agent Framework package on November 17 while a malicious release was live.
- Build servers and CI pipelines that pull the latest version without pinning one, and ran during that window.
- Container images built during that window, which may still contain the bad version.
- Contractors and agencies that build software for you. Their machines may hold your Azure keys.
If every environment was pinned to 1.26.0 or earlier and nothing pulled a newer version that day, the sources give no reason to think you were exposed.
What to do this week: Agent Framework PyPI checklist
| Step | What to do | How to check |
|---|---|---|
| 1. Find it | Search every project for Microsoft's Agent Framework package | Check requirements files, lockfiles and container images for 1.26.1 or any version above 1.26.0; run pip list in each environment |
| 2. Check timing | Look for installs or builds on November 17, 2026 | Review CI and build logs from around 20:35 to 21:30 UTC |
| 3. Pin the safe version | Set the package to 1.26.0 | Pin the exact version in requirements and lockfiles, then rebuild |
| 4. Rotate Azure secrets | Replace Azure and Azure DevOps credentials on any machine or pipeline that ran the bad version | Revoke the old keys and tokens, don't just add new ones |
| 5. Rotate everything else | Replace other secrets kept in environment variables and local config files on those machines | API keys, database passwords, third-party tokens |
| 6. Review activity | Look for unfamiliar sign-ins or changes in Azure and Azure DevOps since November 17 | New users, new tokens, pipeline edits, unexpected deployments |
| 7. Clean caches | Remove cached copies of the bad version | Internal package mirrors, build caches, saved images |
| 8. Rebuild if you can | Our sources do not say whether the stealer left anything behind | Rebuild affected machines or containers from a clean image rather than only uninstalling |
Uninstalling the package does not get back secrets that were already taken. The rotation steps matter most.
Why this matters for a company network or website
A developer's laptop rarely holds just one key. One Azure DevOps credential can reach source code, build pipelines and the systems that deploy a website or app to customers. An attacker holding that credential can push changes that look like ordinary work.
This happened under a trusted name. The package was Microsoft's official release, and teams that keep up with updates were the most likely to pull it. Being careful about where a package comes from would not have caught this one; pinned versions, short-lived credentials and quick key rotation would have limited the damage.
If software is built for you by an agency or freelancer, ask them this week whether any of their environments pulled 1.26.1, and whether your Azure keys were on those machines. We cover who owns that job in our posts Who Is Patching the Software That Runs Your Business? and AI Can Build Your App. Who Is Going to Keep It Running?
Microsoft Agent Framework PyPI malware questions people are asking
Which version of Microsoft Agent Framework is malicious?
Version 1.26.1. Wiz says there were two malicious versions, but the second version number is not in our sources. Treat any version above 1.26.0 published on November 17, 2026 as suspect.
What is the safe version of Agent Framework?
Wiz says: "The most recent safe version is 1.26.0." Pin to that version until Microsoft publishes its own guidance.
Is the malicious package still on PyPI?
No. According to Wiz, both versions were no longer available on PyPI as of 21:30 UTC on November 17, 2026. Copies can still sit in mirrors, caches and container images.
What did the Agent Framework malware steal?
Secrets from environment variables and local configuration files, with a focus on Azure and Azure DevOps credentials.
How did hackers get into Microsoft's PyPI project?
It is not known. Wiz says the attacker appears to have compromised a client certificate and private key for Microsoft ESRP. Microsoft has not given an explanation in the sources we used.
What is Cairn?
Cairn is the name the malicious tooling uses for itself. Wiz describes it as an open-source AI penetration-testing framework. That name does not identify the attacker.
How many people installed the bad version?
Not known. The package gets roughly 200,000 monthly downloads (BleepingComputer), or 192,534 (OX Security). Nobody has published how many of those were the malicious releases.
I removed 1.26.1. Am I safe?
Removing it stops it from running again. Any secrets it already collected are still exposed, so rotate them.
The bigger lesson
A trusted publisher name is not a guarantee. An official package from one of the largest software companies carried a credential stealer for less than an hour, and that was enough to put cloud keys at risk for anyone who updated at the wrong moment. Pin exact versions, keep long-lived secrets off developer machines where you can, and know in advance who rotates your keys when something like this happens. Being up to date and being safe are different things, as we explained in our post Updated Is Not the Same as Patched.
To have your build pipeline and cloud credentials reviewed, see security hardening or request a quote.
Get security alerts like this by email, with our free checklist, at security alerts.
Sources: Wiz; OX Security; BleepingComputer.
Have a question about this article?
Ask it here and we will answer you directly by email. Questions are private, never posted.
Security alerts by email
Hear about the next one first.
When attackers start using a flaw in software businesses run on, we email what it is, whether it touches you and what to do that week, with a round-up of our latest security writing. Free, one-click unsubscribe. You get the checklist straight away: 12 things to secure this week.
Want a result like this?
Security alerts, breaches and the technology a business runs on: what happened, who it affects and what to do.
Get a Tech QuoteKeep reading
ARTEX AI Hacking Framework: Storm-0940 Tried Sign-Ins on More Than 2,250 Tenants and Reached Privileged Access in Under 3 Hours, What Is Confirmed and What to Check
Microsoft tied a campaign to Storm-0940, which tried sign-ins on more than 2,250 tenants and in some cases rea...
Cursor Security Incident: Malicious Claude Code Extension Versions 2.1.312 and 2.1.313 Tried to Plant an SSH Backdoor, Credentials Exposed, and What to Check This Week
Cursor confirms an attacker obtained an access key, malicious Claude Code extension versions 2.1.312 and 2.1.3...
Puppet Enterprise Vulnerability CVE-2026-115731: CISA Lists the Exploited Flaw, the Certificate Key Risk, and the Fixed Versions to Install
CISA added Puppet Enterprise flaw CVE-2026-115731 (CVSS 9.4) to its exploited list on November 17. Here is wha...