CVE-2023-6019: Ray dashboard cpu_profile parameter OS command injection

Automated analysis

Generated by VULONE's analysis model from the NVD record, CISA KEV and EPSS data on 18 September 2026. Confidence: high.

critical priorityUnauthenticated remote OS command injection with CVSS 9.8 and very high EPSS, though not yet in KEV.

What it is

Ray's dashboard exposes a cpu_profile URL parameter that is passed to an OS command without sanitization, allowing command injection. The flaw is fixed in Ray 2.8.1 and later, and it matters because the dashboard is remotely reachable and the injection requires no authentication.

Impact

An unauthenticated remote attacker can execute arbitrary OS commands on the host running the Ray dashboard, leading to full compromise of that system and any data or credentials it holds.

Attack surface

Reached over the network via the Ray dashboard's cpu_profile URL parameter; the CVSS vector shows no privileges or user interaction required, so any host that can reach the dashboard can attempt it.

Exploitation

CISA KEV does not list it, but EPSS is 0.7463 (99.5th percentile) and the only references are tagged Exploit, indicating public exploit material exists and exploitation is likely.

What to do

• Upgrade Ray to 2.8.1 or later as the primary fix.
• If immediate upgrade is not possible, restrict network access to the Ray dashboard to trusted hosts only.
• Place the dashboard behind an authenticating reverse proxy or VPN rather than exposing it directly.
• Run the Ray dashboard process with a least-privilege account and isolate it from sensitive systems.
• Monitor for unexpected child processes spawned by the dashboard service.

Detection

• Inspect HTTP request logs for cpu_profile parameters containing shell metacharacters such as ;, |, $(), or backticks.
• Alert on child processes spawned by the Ray dashboard process that are not expected Ray components.
• Monitor outbound network connections from dashboard hosts for signs of post-exploitation activity.
• Search for known exploit payload strings in dashboard access logs.

This assessment is produced automatically and is not human-reviewed. Verify against the vendor advisory before acting on it.