Hackers are using Artex and other agentic AI-based frameworks to scan public GitHub repositories for exposed credentials and API keys, according to a report Thursday from Microsoft Threat Intelligence.
The credentials are then deployed in downstream attacks against targeted organizations. Starting Oct. 28, researchers found a single IP address using compromised but still-valid application credentials to sign into downstream tenants.
In one case, hackers were able to steal credentials and exfiltrate data from Azure Key Vaults and Storage Accounts, as well as abuse Azure AI services. Researchers also saw hackers using embedded shell commands in a user agent to exploit vulnerabilities as well as attempts to steal App Service authentication tokens.
Microsoft researchers said the threat activity overlaps with a campaign using the same IP address, as outlined in research released Wednesday by Todyl. That report showed how hackers with stolen secrets for Entra ID signed into Microsoft 365 and Azure environments at various managed service providers.
The threat activity follows a series of attacks where hackers exploited Artex and other agentic AI tooling to automate their attacks. Artex is a tool that traditionally has been used for penetration testing, which involves hacking an internal IT environment to check for vulnerabilities that could be exploited in a real attack.
Global targeting
Artex was previously used by a China-nexus threat actor to abuse ConnectWise Cloud Backup. The attacker, tracked as Storm-0940, targeted downstream Microsoft 365 tenants.
The same actor was linked to attacks against South Korean banks and dozens of Japanese companies. It is not immediately clear who is behind the scanning of GitHub repositories.
CrowdStrike released a report earlier this week outlining how Artex was assisted in multiple espionage attacks across the globe. More than 200 organizations have been targeted globally since mid-October.
Hackers have applied various agentic frameworks to scan for GitHub secrets since August, according to Microsoft researchers. The queries have asked for the names of targeted organizations, the keyword for secrets, the keyword that is linked to cloud services as well as keywords related to other AI service providers.
If an organization finds their secrets exposed in a public repository, they should assume those secrets have been compromised, according to Microsoft researchers.
