Skip to content
LIVE
The Executives BriefThe Executives Brief

Hijackers shipped malware through Microsoft's own PyPI account, twice

Two poisoned versions of Microsoft's agent-framework, downloaded nearly 200,000 times a month, stole Azure, GitHub and Docker credentials while Nadella headlined Ignite.

ByLama Al-RashidTechnology Correspondent, The Executives Brief
·4 min read
Hijackers shipped malware through Microsoft's own PyPI account, twice
Executive summary

Microsoft's official agent-framework for Python was hijacked, with versions 1.26.1 and 1.26.2 pushed through the company's own PyPI publisher account to steal developer credentials while CEO Satya Nadella headlined Ignite. For any team running AI build pipelines, the incident proves that a trusted publisher account, not a lookalike package, is now the fastest route into your cloud credentials.

Someone pushed malware through Microsoft's own PyPI publishing account, and they did it twice. Security researchers at OX Security and OpenSourceMalware flagged versions 1.26.1 and 1.26.2 of Microsoft's agent-framework for Python, the company's official toolkit for building and deploying AI agents, as malicious. The package pulls nearly 200,000 downloads a month. The timing is brutal: CEO Satya Nadella was on stage at Ignite, Microsoft's annual customer and investor showcase, while the open source team back in Redmond was dealing with what appears to be unauthorized access to an administrator account on Microsoft's GitHub organization.

What those two versions do is worse than a bad release note. According to the researchers, the infected package harvests environment variables, credential files, and other secrets tied to Azure, Git, Docker, Hugging Face, and other sensitive developer services. It never needs to be imported. Simply having it installed is enough: any Python interpreter that starts up in that environment triggers the payload. To keep up appearances, the malicious build still pulls the genuine 1.26.0 version of agent-framework so the code runs normally. In the background, it encrypts what it collects into a base64-encoded chunk of JSON and ships it to a hardcoded webhook.site address.

The nastier detail is that the malware does not wait for secrets to drift past. It goes and asks for new ones. Researchers say it requests fresh identity tokens from GitHub Actions, Azure DevOps, and Azure itself, then exfiltrates those to the same webhook. OpenSourceMalware put the risk plainly: "That combination of passive theft and active token requests makes it especially dangerous in cloud build environments, because even short-lived tokens are useful when they're exfiltrated right away." That is the sentence to read twice if you run CI/CD pipelines. The industry's standard defense against credential theft is to make tokens short-lived so a stolen one expires before it is useful. This malware collapses that window to seconds.

OX Security's timeline is the part that turns an incident into a governance problem. Version 1.26.1 was reverted to a safe version. The attacker then uploaded 1.26.2 through Microsoft's official PyPI publisher account, which suggests that, at least at the time of the report, the credentials had not been locked down. Read that again: the fix did not hold. For anyone who treats a vendor's official publisher account as a trust anchor, this is the uncomfortable lesson. The compromise was not a lookalike package with a cleverly misspelled name. It came through the front door, signed by the account Microsoft uses to ship its own code. Every lockfile, every internal mirror, and every software bill of materials that inherited agent-framework inherited the problem with it.

The blast radius is still being mapped. The agent-framework GitHub repo was offline as of writing. So was Microsoft's repo for ONNX Runtime, an AI accelerator. Other Microsoft repos remained available, and neither Microsoft nor the security researchers have linked the ONNX pull to the malware campaign. Microsoft did not immediately respond to questions. The silence is easy to explain: the PR machine was in San Francisco for Ignite, and the rank and file have been putting out other fires. But for a company selling AI infrastructure to the enterprise, an unanswered supply chain question during its biggest customer event is its own kind of damage.

Zoom out and the structural problem is bigger than one package. Public registries like PyPI are built on a trust model that assumes the account behind a release is the account that should be releasing it. That model has held up remarkably well, but it concentrates enormous power in a single set of credentials. Microsoft's agent-framework sits in a particularly sensitive spot because it is a framework, not an app. Developers install it to build and deploy AI agents, which means it tends to live inside build environments where cloud tokens, container registries, and model-hub keys all sit within reach. A framework that phones home is a framework that can phone home with everything around it.

The practical response for any team that touched these versions is not subtle. Treat the environment as compromised, rotate Azure, GitHub, Docker, and Hugging Face credentials, and check what the webhook received. The malware is not persistent, running only as long as the malicious version stays in a site-packages directory, but that is cold comfort when the exfiltration already happened. The harder question is structural: who inside your organization can publish to a package registry, and are those accounts protected the way you protect production? For boards, this is another reminder this year that the software supply chain is a governance surface, not an IT chore. The next hijack will not announce itself either.

Executive ActionsLocked

This story's Key Insights and Take-aways are locked.

Create a free account to unlock Executive Actions for one credit.

Register to Unlock

Always free for Executives Club members. Join the Club

More in Technology