What happened
On Monday, August 17, the Cybersecurity and Infrastructure Security Agency added CVE-2025-62593 to its Known Exploited Vulnerabilities catalog, citing evidence of active attacks. The flaw sits in Ray, a widely used open-source framework for running AI and Python workloads across clusters of machines. Rated 9.4 out of 10 in severity, it stems from missing authentication on critical endpoints such as the jobs API, which lets attackers execute code on exposed servers. Researchers say the bug can even be triggered through a victim's browser using DNS rebinding, so a developer running Ray locally can be compromised just by visiting a malicious web page. Operators of the RondoDox DDoS botnet folded the exploit into their toolkit almost immediately, and a separate campaign dubbed ShadowRay 2.0 has been hijacking unpatched Ray clusters to mine cryptocurrency on expensive GPU hardware. The fix shipped in Ray version 2.52.0, and CISA gave federal civilian agencies an unusually short deadline of August 20 to remediate.
Why it matters for your business
Ray is not just a big-company tool. Plenty of small teams experimenting with AI, data pipelines, or machine learning use it on a cloud server or an office workstation, often with default settings and no authentication. A three-day federal deadline is CISA's way of saying this is being exploited broadly right now, not theoretically. If your business or your contracted developers run AI workloads, this is the kind of quiet infrastructure component that never shows up in an inventory until it becomes the way someone gets in.
What to do about it
- Ask whoever manages your servers or AI projects whether Ray is in use anywhere, including test machines
- Upgrade any Ray installation to version 2.52.0 or later
- Keep Ray dashboards and APIs off the public internet and behind authentication
- If an exposed instance is found, check it for unfamiliar jobs or crypto-mining activity before assuming it is clean