'Zealot' Shows What AI's Capable of in Staged Cloud Attack

Unit 42 built an autonomous multi-agent system called "Zealot" that, given a single natural-language prompt, autonomously mapped a misconfigured Google Cloud environment, exploited a server-side request forgery to obtain a service account token, and exported BigQuery data to a storage bucket it then read — completing the attack in two to three minutes; the report warns that agentic AI acts as a force multiplier for exploiting existing cloud misconfigurations and recommends automation and rapid remediation for defenders.