Code Defence Cyber security

CISA adds Ray AI framework code injection vulnerability CVE-2025-62593 to KEV catalog

A critical code injection vulnerability residing within a dominant open source distributed artificial intelligence compute framework has been officially added to the federal directory of active cyber threats. The defect permits unauthenticated remote actors to submit malformed compute jobs to execute arbitrary system commands across underlying GPU cluster nodes.

Tracked as CVE-2025-62593, the vulnerability impacts default installations of the Ray AI framework where cluster management ports are exposed without authentication controls. Threat actors transmit malformed job submission payloads to force worker nodes to execute shell commands, granting full host authority over cluster instances. Due to sustained wild target scanning across public cloud infrastructures, CISA added the flaw to the Known Exploited Vulnerabilities catalog under Binding Operational Directive 26-04 mandates.

Subverting a centralized artificial intelligence cluster manager introduces severe supply chain and cloud hosting risks. Because AI training nodes hold database access credentials, proprietary model weights, and cloud service integration keys, an unauthenticated host takeover allows threat networks to steal proprietary model intellectual property, access customer datasets, and pivot into corporate cloud environments.

– Restrict network visibility of Ray dashboard and job submission ports, gating management endpoints behind zero trust proxies.

– Apply security updates and authentication enforcement patches distributed by the Ray-Project development team.

– Inspect cluster execution logs for anomalous shell command executions originating from Ray worker worker processes.

– Rotate all cloud infrastructure credentials and API keys stored within AI cluster environment variables.

Artificial intelligence compute plane protection demands strict interface isolation and continuous job submission auditing to ensure distributed execution clusters remain completely protected from unauthenticated code injection payloads. #CodeDefence #RayProject #AISecurity #RCE #CodeInjection #CISA #KEV #CloudSecurity
/

Scroll to Top