ARTEX Scans GitHub Repositories and Exfiltrates Azure Key Vault Secrets

EXECUTIVE SUMMARY:

Threat actors have recently leveraged an agentic AI offensive framework to scan public code repositories for exposed secrets, including application credentials and API keys. This automated approach accelerates the discovery of authentication material that was historically limited by the time required for manual review. The initial access vector relies on publicly available data, where developers inadvertently commit sensitive information to open-source platforms. Once these secrets are identified, the framework uses them to gain unauthorized access to cloud environments, posing a significant risk to organizations that store credentials in code or configuration files without adequate protection.

The mechanism involves the AI framework submitting large volumes of queries to locate target-specific secrets, then using valid leaked credentials to perform service principal sign-ins across multiple tenants. Upon successful authentication, the agent conducts cloud resource discovery, enumerating subscriptions, resource groups, and permissions to map the environment. It targets operations that return storage account keys, connection strings, and machine learning datastores, while also accessing SharePoint and OneDrive for file downloads. The agent demonstrates adaptive behavior by adjusting API versions and HTTP methods after failed requests to ensure successful data retrieval. In one observed instance, the system progressed from scanning repositories to exfiltrating data from key vaults in just five minutes, utilizing compromised OAuth applications to execute scripts on virtual machines for further lateral movement and persistence.[emaillocker id="1283"]

This activity illustrates a growing trend of threat actors using agentic AI to accelerate all stages of the attack chain, reducing the need for hands-on-keyboard interactions. The campaign has impacted multiple customers, with compromised multitenant application credentials used to access downstream tenants and abuse AI services. The rapid progression from initial discovery to data exfiltration highlights the urgency of automated defensive measures. Organizations must immediately rotate exposed credentials, enforce least privilege for workload identities, and migrate away from secret-based authentication to managed identities. Additionally, implementing behavioral detection for anomalous API usage and deploying defensive honeypots can help mitigate the impact of these autonomous attacks.

THREAT PROFILE:

MBC MAPPING:

REFERENCES:

The following reports contain further technical details:

• https://www.microsoft.com/en-us/security/blog/2026/11/05/threat-actors-use-agentic-ai-tooling-to-find-and-abuse-exposed-secrets/