A toolkit that still ran its real features is exactly how a supply-chain hit hides in plain sight. According to BleepingComputer, attackers published malicious versions 1.26.1 and 1.26.2 of Microsoft’s official agent-framework package on PyPI, each built to steal cloud credentials and other sensitive material from developer systems. Both releases were quarantined within about an hour, yet anyone who installed during that window is told to treat local secrets as exposed and rotate them. How the publisher account was abused is still not established in that reporting, and no public count of hit developer environments has been released.
Two malicious builds under Microsoft's PyPI name
Clean agent-framework 1.26.0 was the last good line before the bad tags. According to BleepingComputer’s timeline, version 1.26.1 appeared on PyPI on 17 November 2026, then was quarantined; the attacker pushed 1.26.2 roughly two and a half minutes after that first quarantine, and 1.26.2 itself was quarantined about eight minutes later. The whole malicious window sat inside roughly an hour, not a multi-day unnoticed dwell.
Trust in the package name did the heavy lifting. The project sat on Microsoft’s official PyPI namespace, so pip install looked routine for teams already pulling agent-framework for agent and orchestration work. Monthly download volume near 200,000 meant the blast radius was not a niche hobby package. BleepingComputer reported that Microsoft had not explained, by the time of that article, how publishing rights for those releases were obtained. That gap remains open in the public record cited here: the method used to obtain Microsoft PyPI publishing rights is not documented in the primary write-up.
Malicious builds kept a legitimate agent-framework-core dependency in place. Developers who upgraded could see normal framework behavior while the extra payload armed itself. Adjacent developer supply-chain cases flagged in the same reporting cluster, including a LiteLLM release-pipeline info-stealer incident and OVHcloud CLI GitHub backdoors aimed at developer secrets, show the same industry pattern without proving a shared operator for this Microsoft package event.
What the .pth hook took from developer machines
Python’s .pth startup mechanism was the quiet delivery path. Once the bad wheel or sdist landed, a .pth file caused stealer code to run at interpreter startup, including in environments that never imported agent-framework explicitly. No extra “run the demo” step was required for the hook to fire.
Harvest logic went after secrets already present on the machine. Environment variables, local configuration files, and other processes (where permissions allowed) were in scope. Targeted cloud and developer services named in the public analyses included Azure, AWS, GitHub, Docker, Hugging Face, and OpenAI-related material. The payload tried to obtain Azure managed-identity access tokens and Azure DevOps OIDC tokens. Version 1.26.2 added GitHub Actions OIDC token requests and searches under GitHub runner directories. Collected data left for an attacker-controlled webhook.site endpoint.
Independent reverse-engineering notes from firms such as Socket, StepSecurity, OX Security, and Wiz lined up on that behavior. None of the cited public reporting confirms that stolen managed-identity or OIDC tokens were later used successfully against live cloud tenants. Volume of credentials exfiltrated is also unpublished. Named threat-actor attribution is absent from the BleepingComputer account.
| Control surface | What showed up here | Why it did not save installs |
|---|---|---|
| PyPI project trust | Official Microsoft package name | Publishing path was already abused |
| Package behavior | Real agent-framework-core still loaded | UI and imports looked healthy |
| Runtime secrets | Env, configs, MI/OIDC attempts | Secrets sat in the dev process space |
| Registry response | Quarantine in about an hour | Installs inside the window still ran |
Interactive workforce login was not the story the sources tell. No MFA method is named anywhere in the incident path. Stolen material was authorization data and tokens read from developer runtimes after a successful package install, not a password or one-time code typed into a fake corporate login page.
Quarantine speed versus rotation reality
PyPI quarantine stopped new victims from pulling 1.26.1 and 1.26.2 through the index. It did not unscrew machines that had already installed those builds, cached wheels, or mirrored internal indexes that had fetched the bad versions. BleepingComputer’s practical guidance was blunt: isolate affected environments and rotate any credentials that the compromised environment could reach.
Rotation lists write themselves from the stealer’s shopping list. Cloud access keys and tokens, CI OIDC material, Docker and GitHub credentials, and anything sitting in env files next to agent experiments all belong on the burn pile if that host ran the malicious build. Short registry response time cuts the install graph; it does not prove zero successful exfiltration for early adopters.
Developer laptops and build agents remain high-value because they already hold long-lived cloud identities for shipping code. A supply-chain stealer that rides a trusted package skips the need for a phishing attack against a human at a sign-in form. The same class of outcome shows up when release pipelines or official CLIs are backdoored in other vendor ecosystems: the install is the access, and the secret store is local memory and disk.
Microsoft’s agent-framework event is therefore a publishing-and-runtime problem first. Corporate SSO hardening would not have rewritten the .pth file out of a bad wheel. Operators still need package allowlists, pinned hashes, internal indexes that do not blindly float latest tags, and rapid credential invalidation when a signed name on a public registry turns hostile for an hour.
If you want the authentication-side reading on what device-bound login can and cannot change once a supply-chain stealer is already executing, read the related article on mfa2point0.com: Microsoft PyPI stealer and MFA limits on supply-chain tokens.
FAQ
How did the Microsoft agent-framework PyPI compromise start?
The Microsoft agent-framework PyPI compromise started when attackers published malicious versions 1.26.1 and 1.26.2 under Microsoft’s official agent-framework project on PyPI on 17 November 2026. According to BleepingComputer, both builds were quarantined within about an hour, but the report does not establish how publishing access was obtained.
What did the malicious agent-framework packages do on install?
The malicious agent-framework packages installed a Python .pth startup hook that ran a credential stealer automatically, even when developers did not import the framework. The stealer collected secrets from environment variables and configs, tried to obtain Azure managed-identity and DevOps OIDC tokens (with GitHub Actions token logic in 1.26.2), and sent data to a webhook.site endpoint while still loading legitimate agent-framework-core so the toolkit appeared to work.
Were Azure or GitHub tokens confirmed used after theft?
No public confirmation in the BleepingComputer reporting shows successful use of stolen Azure managed-identity, Azure DevOps OIDC, or GitHub Actions tokens from this incident. The payload attempted those requests; outcome after exfiltration is not documented in that source.
Is this the same as phishing employees for Microsoft 365 MFA codes?
No. The Microsoft agent-framework PyPI incident, as reported, is a supply-chain package compromise that ran malware inside developer environments after install. Sources do not describe a spoofed workforce login, helpdesk reset, or any named MFA factor in the attack path.
What should teams that installed 1.26.1 or 1.26.2 do first?
Teams that installed agent-framework 1.26.1 or 1.26.2 should isolate those environments and rotate credentials that the host could read, matching BleepingComputer’s guidance that any potentially exposed secrets in reach of the compromised environment should be rotated immediately. Pin back to known-clean 1.26.0 or a later verified good release only after the local secret sprawl is treated as burned.