Google Threat Intelligence Group (GTIG) linked an April 2026 intrusion to an exposed GitHub personal access token, connecting the credential to unauthorized AI workloads and high-performance computing in Google Cloud.
The group said the activity included enabling Gemini Enterprise and creating a service account with Editor-level permissions. LiteLLM and Manus containers ran on Cloud Run services accessible to the public.
GTIG said the attacker requested higher quota for NVIDIA RTX 6000 hardware, then launched additional 48-vCPU compute instances.
LLMjacking describes using an organization’s cloud computing capacity for AI workloads without authorization.
