Attackers Are Turning AI Agents Into Automated Intrusion Tools, Google Warns

Google Threat Intelligence Group (GTIG) and Mandiant have seen multiple threat actors abusing open-source agentic pentesting frameworks called ARTEX, Hexstrike and CAIRN. The tools were observed during incident response engagements, where attackers used them across several phases of the attack life cycle. A recent Wiz report points to similar activity elsewhere in the threat landscape.

GTIG says this marks a shift from basic prompting of AI models to agentic workflows and LLM-enabled automation. The frameworks can chain complex attack paths together automatically, which lets intrusions run at machine speed and scale far faster than before. The researchers also note that attackers may be building custom agentic tools of their own.

The methods themselves have not changed. Attackers still rely on compromised credentials and unpatched systems. What is new is how quickly and widely automated tools can exploit these known weaknesses. Mandiant Consulting recommends long-standing best practices: control-plane isolation, short-lived credentials, least privilege, segmentation and behavioural detection. This article covers only the opening of the original report, so the full research may include further detail.