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

IncidentAI infrastructure

ShadowRay 2.0: exposed Ray AI clusters turned into a botnet

CriticalAnyscale Ray

A new campaign abused Ray's unauthenticated Jobs API to take over internet-exposed AI clusters, spread the payload to every node using Ray's own scheduler, mine cryptocurrency, steal data and launch DDoS attacks.

Replay the attack step by step in the Learning Center

How it happened

  1. Attackers found Ray servers reachable from the internet and submitted jobs through the unauthenticated Jobs API.
  2. The jobs pulled multi-stage payloads that researchers say show signs of AI-generated code.
  3. Ray's own scheduling pushed the payload onto every node, making the botnet self-spreading.
  4. They mined cryptocurrency quietly, killed rival miners, kept access through cron and systemd, and stole models, code and credentials.

Impact

Attackers got expensive GPU compute plus the models, code and credentials on the cluster.

The fix

The CVE remains unpatched and disputed; run Ray only on trusted networks with firewall rules and authentication on the dashboard port.

Lessons

  • 'By design' flaws still get exploited; scanners that skip them leave blind spots.
  • Never expose AI orchestration dashboards or job APIs to the internet.
  • Watch GPU clusters for resource spikes, new cron entries and unexpected outbound connections.

Sources