Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

IncidentAI infrastructure

ShadowRay: AI clusters hijacked through an unauthenticated jobs API

CriticalAnyscale Ray

Attackers took over internet-exposed Ray clusters, a widely used framework for scaling AI, through a job-submission API with no authentication, then mined cryptocurrency and stole credentials.

Replay the attack step by step in the Learning Center

How it happened

  1. Ray's Jobs API has no authorization, and many clusters were reachable from the internet.
  2. Attackers submitted jobs that ran arbitrary commands.
  3. They installed cryptominers and reverse shells.
  4. They harvested OpenAI, Hugging Face, Stripe and cloud credentials from the cluster.

Impact

Costly GPU clusters, cloud credentials and models could be taken over.

The fix

No patch; the vendor says Ray must run on strictly controlled networks, and Ray now documents an optional token authentication mode.

Lessons

  • Never expose AI orchestration dashboards or job APIs to the internet.
  • Keep long-lived secrets out of cluster environments.
  • Watch GPU clusters for unexpected processes.

Sources