AI agent tool went from GitHub secrets to Azure vault in 5 minutes

Microsoft Threat Intelligence tells that since August 2026 attackers use an agentic AI framework, ARTEX, to find secrets in public GitHub code. These are credentials and keys for AI services. In one example, an agent went from the scan to data theft from Azure Key Vault in 5 minutes. The report names no victim.

How it works

The agent sends many queries to GitHub. They frequently have the name of a target and terms for secrets or cloud services. Then it uses the keys that it finds to sign in, find the parts of the cloud and get data. A person does less work.

The numbers

From October 28, 2026, Microsoft saw sign-ins with correct application credentials on many applications in some tenants (customer accounts). The attacker took data from Key Vaults, Storage Accounts, SharePoint and OneDrive.

What is not known

Microsoft does not tell how many organizations are hit. One address is a Tor exit node, which can come from attacks that are not related. Todyl told of a campaign that is almost the same.

Insight: Attack speed beats the response time of people

In October 2026, Google and Mandiant saw one group go from a first account to deep access in 4 hours. Exaforce tells that attackers get data in 15 minutes or less. Microsoft tells of 5 minutes. The loop of the attacker is faster than the inspection loop of the defender, and the side that uses automatic response gets the advantage. Microsoft also tells that defenders must use AI.