Criminals set AI agents loose on GitHub to hunt for leaked keys and get into cloud accounts within minutes

Microsoft Threat Intelligence warns that attackers are using ARTEX, an offensive security framework built on AI agents, to search public GitHub repositories for leaked passwords and keys. In one case, an agent went from searching GitHub to siphoning data out of an Azure Key Vault in about five minutes. The agents used the captured credentials to sign in to dozens of applications belonging to different customers.

Microsoft saw the mass searches from August onwards, followed in late October by sign-ins and abuse in cloud environments, including access to SharePoint, OneDrive and mailboxes. When a request failed, the agents adjusted their approach on their own until it worked. They were mainly after keys for AI services.

According to GitGuardian, ARTEX appeared on GitHub in late July and was removed on 8 October, but the attacks are continuing. Censys counted about 1,400 publicly reachable ARTEX installations in early November and Check Point more than 1,800, while TechCrunch also reported hundreds of exposed instances. The campaigns were mostly opportunistic and aimed at cryptocurrency, gaming accounts and certificates used to sign Apple software. Microsoft stresses that a leaked key must be treated as compromised, even if the file or the git history has since been cleaned up.