Wiz reconstructs an AI-directed intrusion crossing cloud, identity and delivery systems

In an investigation published October 19, Wiz CIRT described an attacker who exploited an internet-facing Kafka UI to execute commands, deployed a backdoor and harvested workload secrets. The operation continued in several cloud and platform control planes. Stolen and minted identities allowed AWS Organizations access, GCP service-account token issuance, changes to IdP group membership and creation of Kubernetes workloads through CI/CD and GitOps services.

The distinctive AI attribution rests on high-tempo cross-platform workflows, repeated correction after API failures, code-generation artifacts and coordinated workstreams. The report presents observations from real incidents, but it does not publish signed model transcripts for every action. An originating IP also scanned more than 600 other customer environments: scanning is not evidence of 600 compromises. In the investigated victim environment, the attacker used several channels including CI logs, source control, ConfigMaps and an external request collector for data movement. Some routes were blocked or attempted, so total transferred bytes and confirmed losses remain unknown.

The actual foothold and privilege transitions are better established than the precise degree of model autonomy. Wiz describes compromise of an exposed Kafka interface, then abuse of IdP administration, cloud credentials, service accounts and direct GitOps control-plane mutations. A deployed manifest could reach production without an attacker-authored Git commit because an already compromised controller credential could change state directly. The more than 600 additional environments were scanned, not reported compromised.