Description

CISA has added CVE-2025-62593, a critical vulnerability in the open-source Ray AI/ML computing framework, to its Known Exploited Vulnerabilities (KEV) catalog after evidence of active exploitation. The flaw carries a CVSS 4.0 score of 9.4 and can enable remote code execution through a victim’s web browser. The vulnerability affects Ray versions prior to 2.52.0 and stems from insufficient protection of browser-accessible Ray dashboard functionality. Ray historically relied on a browser User-Agent check to distinguish potentially malicious requests, but this control can be bypassed because browser requests can modify the User-Agent header. An attacker can combine this weakness with DNS rebinding to make a victim’s browser interact with a locally running Ray instance. The attack requires the victim to be running Ray and subsequently visiting a malicious website or encountering malicious advertising. Under vulnerable conditions, the browser can act as an intermediary to reach Ray endpoints and trigger arbitrary code execution on the developer’s system. The technique may also expose Ray instances located inside private corporate networks.