LLMjacking Campaigns Target Google Cloud Vertex AI Credentials, Exceed $100K Daily in Fraudulent Charges

Threat actors are executing LLMjacking campaigns to compromise Google Cloud Vertex AI API keys and service account credentials. Initial access vectors include phishing, hardcoded keys in leaked source repositories, misconfigured cloud storage, exposed environment variables, and client-side code exposure. Attackers fraudulently invoke Vertex AI LLM APIs to consume GPU/TPU compute resources. Mandiant investigated 2026 incidents exceeding $100,000 in daily fraudulent cloud charges, with total bills approaching $1 million per victim organization.

• Incident Overview: Vertex AI Credential Hijacking

  • Google Threat Intelligence Group and Mandiant investigated multiple LLMjacking incidents in 2026.
  • Attackers targeted Google Cloud Vertex AI API keys and service account credentials.
  • Unauthorized LLM API invocation produced over $100,000 in fraudulent charges daily.
  • Total cloud bills per incident reached up to $1 million.

• Attack Vector & Campaign Mechanics

  • Phishing messages and credential harvesting pages captured Vertex AI credentials.
  • Leaked source code repositories contained hardcoded API keys.
  • Misconfigured cloud storage buckets and environment variables exposed secrets.
  • Exposed client-side application code enabled direct credential extraction.
  • Stolen credentials enabled remote inference without enterprise authorization.

• Threat Group Profile & Scale of Impact

  • Financially motivated threat actors conducted large-scale LLMjacking campaigns.
  • Enterprise victims experienced unauthorized enterprise AI model usage.
  • Fraudulent compute resource consumption targeted expensive Vertex AI infrastructure.
  • Financial loss ranged from $100K per day to $1M per incident.

• Indicators of Compromise & Defensive Actions

  • Review unauthorized LLM API invocation logs for anomalous activity.
  • Audit fraudulent compute resource consumption records for unexpected spikes.
  • Rotate Vertex AI API keys and service account credentials upon detection.
  • Eliminate hardcoded keys from repositories and client-side application code.
  • Implement anomaly detection on Vertex AI API call volumes.
  • Enforce least-privilege IAM policies and secure environment variable storage.

• Conclusion

  • LLMjacking is an evolution of cloud resource hijacking into the AI and LLM era.
  • Google Cloud Security has issued explicit warnings about Vertex AI credential abuse.
  • Security teams must treat LLM API keys as high-value secrets equivalent to root credentials.
  • Continuous credential hygiene and real-time API monitoring are critical to prevent six-figure cloud fraud.