Google Ties April Cloud Intrusion to Unauthorized 48-vCPU Instances

Google Threat Intelligence Group identified an April 2026 cloud intrusion that began with an exposed GitHub personal access token and expanded into unauthorized AI workloads and high-performance computing.

The attacker enabled Gemini Enterprise and provisioned compute. The intrusion also placed LiteLLM and Manus agent-framework containers on Cloud Run services that were publicly accessible. The activity also involved creating an Editor-level service account.

The attacker requested quota increases for NVIDIA RTX 6000 hardware before launching additional 48-vCPU instances. LLMjacking describes the hijacking of enterprise cloud infrastructure to run unauthorized AI workloads.

Demand to buy AI-related accounts and the number of sellers advertising them rose year over year. Average underground-market prices per account more than doubled in 2026.