Ray, an AI compute engine, has a notable vulnerability in its dashboard HTTP server. The server blocks browser‑origin POST and PUT requests but leaves DELETE unprotected, and the critical DELETE endpoints are unauthenticated out of the box. When a dashboard or agent instance is publicly reachable, an attacker can trigger these DELETE calls from a browser, causing the Serve component to shut down or removing jobs from the system without any user interaction.

All Ray deployments that expose the dashboard host to 0.0.0.0 or otherwise allow network access and run version 2.53.0 or earlier are affected. The issue is corrected in Ray version 2.54.0 and higher, which implements authentication for delete endpoints and restricts access to the dashboard host.

The CVSS score of 5.9 marks this flaw as medium severity, while the EPSS score of less than 1 % indicates a low likelihood of exploitation at present. Attackers who can reach the dashboard—through DNS rebinding, same‑network exposure, or a publicly accessible host—can issue unauthenticated DELETE requests to shut down Serve or delete jobs, leading to a denial of service. This vulnerability is not currently listed in the CISA Known Exploited Vulnerabilities catalog.