Microsoft Agent Framework is a fairly popular tool with roughly 200,000 monthly downloads on PyPI.
Malicious releases 1.26.1 and 1.26.2 published on November 17, 2026
On November 17, 2026, attackers published two malicious releases of Microsoft's official Agent Framework Python package on PyPI: versions 1.26.1 and 1.26.2. Both releases contained a credential stealer that executed automatically during Python startup in affected environments — even when the Agent Framework package was not imported — while retaining the legitimate dependency agent-framework-core 1.26.0 so the framework would appear to function normally as the malware ran in the background.
Multiple security firms discovered and documented the attack, including Socket, Step Security, and Ox Security. Both malicious versions exfiltrated collected data to the same attacker-controlled endpoint: https://webhook[.]site/d381fd80-ec62-42ad-8f5c-d1b15103c9bb.
Credential-stealing behavior and targeted tokens
The first malicious release harvested secrets from environment variables, local configuration files, and other running processes where permissions allowed, targeting credentials and tokens for services including Azure, AWS, GitHub, Docker, Hugging Face, OpenAI, and other services. The malware also attempted to obtain Azure managed identity access tokens and Azure DevOps OIDC tokens, potentially giving attackers access to additional cloud resources.
Version 1.26.2 expanded those collection tactics: it added GitHub Actions OIDC token requests and searches of GitHub runner directories.

Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleAutomatic execution during Python startup: implementation details
According to Socket and Step Security, the first malicious release used agent_framework_meta.pth to execute the credential stealer in agent_framework_meta/_telemetry.py during Python startup. The second release restored the original package initializer and moved the malicious code into af_runtime_diag.pth and af_runtime_diag.py, preserving automatic execution while changing file names and locations.
PyPI response, publishing timeline, and remaining unknowns
PyPI quarantined the first malicious release, but the attacker published a second version roughly 2.5 minutes later, which was quarantined about eight minutes afterward. The rapid re-publication indicates the attacker still had publishing access after the initial quarantine. It remains unclear how the attacker gained access to publish malicious releases under Microsoft's official PyPI project.
BleepingComputer has contacted Microsoft to learn what the point of breach was, but had not received a response by publication time.
What this means for developers who installed the package and for investigators
- Developers who installed either malicious version should isolate affected environments, avoid running Python or pip until those environments are secured, and rebuild from a known-clean environment using the safe 1.26.0 release.
- Any potentially exposed credentials that were within the reach of the compromised environment should be rotated immediately.
- Investigators and incident responders should search egress logs for the specific webhook URL path to determine the exact level of compromise and must distinguish blocked requests from completed transfers to understand whether data reached the attacker-controlled endpoint.
The core unanswered question is straightforward and actionable: how did an attacker obtain the ability to publish malicious releases under Microsoft's official PyPI project? Until that point of access is identified and closed, other projects and users that rely on third-party packages remain exposed to similar risks.
Original reporting: BleepingComputer — Microsoft’s Agent Framework PyPI package compromised to steal credentials




