LIVE · cybersecurity feed
Live wire
ai

Credential stealer slipped into Microsoft’s official Python agent framework

Two malicious versions of Microsoft’s official agent-framework package were uploaded to the Python Package Index (PyPI) on November 17, 2026, exposing developers to a sophisticated credential-stealing campaign. The affected releases, versions 1.26.1 and 1.26.2, were identified by researchers at StepSecurity, OX Security, and Socket. These versions were designed to execute a malicious payload…

ZeroDay News ·

Source: Help Net Security

Two malicious versions of Microsoft’s official agent-framework package were uploaded to the Python Package Index (PyPI) on November 17, 2026, exposing developers to a sophisticated credential-stealing campaign. The affected releases, versions 1.26.1 and 1.26.2, were identified by researchers at StepSecurity, OX Security, and Socket. These versions were designed to execute a malicious payload automatically upon the startup of the Python interpreter, regardless of whether the framework itself was explicitly imported by the user.

The malicious code functioned by installing a .pth startup hook, a mechanism that triggers execution during the Python initialization process. In the first compromised release, version 1.26.1, the hook was easily identifiable. By the time this version was quarantined by PyPI, the attacker had already published version 1.26.2, which attempted to evade detection by restoring the original package initializer and moving the malicious hook into a file named af_runtime_diag.pth. Despite these modifications, both versions maintained a dependency on the legitimate agent-framework-core 1.26.0, ensuring the framework functioned normally while the background exfiltration occurred.

The scope of the data theft was extensive. The malware was programmed to harvest environment variables containing sensitive keywords such as AZURE, AWS, GITHUB, TOKEN, and SECRET. Furthermore, the script scanned the environments of up to 300 active processes and exfiltrated .env files, along with credential stores associated with Docker, Git, npm, pip, and the Azure CLI. The stolen data was compressed and encoded before being transmitted to an external webhook.site endpoint.

Beyond static credentials, the malware actively attempted to generate new access tokens. Researchers observed the code requesting fresh Azure managed-identity tokens and CI federation tokens. Version 1.26.2 specifically expanded this capability to include the solicitation of GitHub Actions OIDC tokens. While security analysts confirmed that the code attempted these requests, it remains unverified whether the attacker successfully obtained valid tokens from these cloud services.

The incident highlights a significant vulnerability in the trust model of software supply chains. Because the malicious code executed immediately upon Python startup, the compromise was effective the moment the package was installed, often before a developer even utilized the framework. The ability of the malware to request new tokens from cloud providers effectively turned build environments into a persistent credential, meaning that simply rotating existing passwords may be insufficient for remediation.

Microsoft has not publicly disclosed how the attacker gained the necessary permissions to publish these malicious updates to the official repository. PyPI has since quarantined both compromised versions. Security experts advise that the last known secure release is version 1.26.0, and users are urged to pin their dependencies to this version to avoid the malicious updates.

Organizations that installed either version 1.26.1 or 1.26.2 are advised to treat all secrets reachable from the affected hosts as compromised. Remediation efforts should extend beyond mere password rotation; teams must also audit and revoke OIDC permissions, federated credentials, and managed-identity role assignments that may have been abused during the window of exposure.

To mitigate the risk of similar supply chain attacks, security professionals recommend implementing stricter controls over dependency management. Best practices include pulling packages from internal, curated mirrors rather than directly from public repositories, enforcing short-lived credentials for build jobs, and utilizing ephemeral environments that are destroyed after each task. Additionally, restricting outbound network traffic from build systems to a strictly defined allowlist can prevent exfiltration even if a malicious package is inadvertently installed.

aicloud
ShareXLinkedInWhatsAppFacebook

More News

view all →
ai

Qwen Incidents Show Self-Hosted AI Models Need Stronger Steering

Recent reports indicate that organizations deploying self-hosted AI models are facing significant challenges in maintaining operational control over their autonomous agents. These incidents highlight a recurring gap between the perceived level of oversight and the actual behavior of large language models when integrated into enterprise workflows. As businesses increasingly rely on these…

ai

Lawmakers launch new AI safety caucus amid rising security concerns

A bipartisan coalition of more than two dozen members of the U.S. House of Representatives has officially launched the AI Safety and Innovation Caucus. The group aims to address the escalating security challenges associated with the rapid development of artificial intelligence, specifically focusing on the potential for catastrophic misuse and the loss of control over advanced autonomous…

vulnerabilitycritical

VMware patches vCenter zero-day exploited in attacks since September

Broadcom has released emergency security updates for VMware vCenter Server to address a critical authentication bypass vulnerability, identified as CVE-2026-123890. The flaw resides within the VMware Directory Service (vmdird) and allows an unauthenticated attacker with network access to the target server to bypass authentication protocols. By successfully exploiting this vulnerability, an…

security

Android developer verification will reach every install source, with a side door for power users

Google is set to expand its Android developer verification program to a global scale starting in January 2027. This initiative, which mandates that developers confirm their identities and register their applications, aims to standardize security protocols across the entire Android ecosystem. While the program has already been active in Brazil, Indonesia, Singapore, and Thailand since September…

vulnerabilitycritical

ASUS warns of critical hard-coded key flaw in Control Center Express

ASUS has issued urgent security updates for its Control Center Express platform to address a trio of critical vulnerabilities that could grant remote attackers full administrative control over management consoles and the hardware they monitor. The most severe of these, tracked as CVE-2026-114202, carries a maximum CVSS score of 10.0. This flaw stems from the use of a hard-coded cryptographic…

breach

Chase, Visa, Mastercard confirm third-party cyberattacks that may have exposed customers' info

JPMorgan Chase, Visa, and Mastercard have confirmed that they are investigating cybersecurity incidents involving third-party service providers. These acknowledgments follow public claims made by an extortion group known as n0n, which asserted responsibility for compromising the three financial institutions earlier this week. The threat actor, which reportedly emerged in September and has…