Ray-Project Ray code injection actively exploited in the wild — added to CISA KEV

CISA added CVE-2025-62593 to its Known Exploited Vulnerabilities catalog on 2026-08-17, confirming in-the-wild exploitation of a code injection flaw in Ray, a widely used open-source distributed computing framework heavily used for AI/ML training and serving workloads. The vulnerability combines CWE-94 and CWE-352 and allows remote code execution.

Ray underlies many production ML training/serving pipelines; confirmed active exploitation combined with reports of GPU clusters being hijacked into a cryptomining botnet (ShadowRay 2.0) demonstrates direct compromise of AI training infrastructure at scale.

Code injection reachable via browser (Firefox/Safari) pivoting to Ray's dashboard/API, exploitable even without direct internet exposure of the Ray service by pivoting through a local/trusted network address; leads to remote code execution on the Ray cluster.

Ray-Project Ray (distributed computing/ML framework) versions prior to 2.52.0

Upgrade to Ray >= 2.52.0; CISA federal due date 2026-08-20; Oligo reports unpatched clusters being converted into a self-replicating cryptomining botnet (ShadowRay 2.0) targeting NVIDIA GPU nodes.