CVE-2025-62593

Ray is a tool developers use to run and manage AI compute. This flaw can let a malicious website trick a browser into bypassing protections so that code runs on the developer’s own computer. Think of it like a “front door” to your development environment that the attacker can disguise using normal-looking web activity.

This matters if your business runs Ray on a machine used by developers (or operators) and that Ray service is reachable on a network interface accessible from users’ browsers. It’s especially relevant for developers who might browse to untrusted websites, because the attack relies on the victim visiting a malicious site. Exploitation is a risk when Ray is exposed/reachable from the network a browser can reach, and the attacker can use DNS rebinding via a malicious website or DNS control.

This is RED because CISA confirmed active exploitation and added CVE-2025-62593 to the KEV catalog. Attackers can drive the compromise through browser-based DNS rebinding, meaning a developer can be targeted simply by visiting a malicious website while Ray is network-reachable.

Confirm exposure

1. Identify where Ray is running (which developer workstations, VMs, and servers).
2. For each machine, confirm the Ray version. If it is earlier than 2.52.0, it’s in scope.
3. Confirm network exposure: check whether the Ray service is reachable from a browser over the network (not just localhost). If it’s accessible from other machines or the internet, treat it as exposed.

Remediate (recommended)

4. Upgrade Ray to the fixed release: ray 2.52.0 (or later).
   • CISA KEV due date: 2026-08-20.
5. After upgrade, validate that the Ray service still functions for your normal workflows and that it is reachable only by intended users.

If you can’t patch immediately

6. Restrict network access so Ray is not reachable from untrusted networks.
   • Allow access only from specific trusted IPs/hosts/VPN networks used by your team.
7. As a temporary risk-reduction step, treat all browsers used on developer machines as potential entry points and reduce exposure until patching is complete.

What to monitor

8. Look for suspicious inbound requests to the Ray service that originate from unexpected domains or sudden spikes tied to browser activity.
9. If you detect suspicious activity, consider isolating the affected developer machine and performing incident response steps (since the impact is arbitrary code execution).

Severity is high: the vulnerability enables remote code execution (arbitrary command execution) on a victim machine. The mechanism is browser/request identification weakness combined with missing protections against DNS rebinding: Ray incorrectly identifies browser requests based on an easily spoofed browser web header, and attackers can modify that header plus use DNS rebinding so the browser treats attacker-controlled content as trusted network endpoints.

Exploitation status: confirmed active exploitation is evidenced by CISA KEV listing (added 2026-08-17) and CISA remediation due date 2026-08-20. There is also press reporting consistent with active exploitation. No public exploit code is noted in the findings, but lack of published code does not reduce the urgency because KEV confirms real-world use.

Fix: Ray is fixed in version 2.52.0 (see the vendor advisory/commit referenced in the patch findings).

Reachability/trigger conditions from findings: the target must be running Ray on a network interface reachable by the user’s browser, and the attacker must control DNS records or a website to perform DNS rebinding; user interaction is required (victim visits a malicious website) but no authentication is required to exploit the vulnerability when conditions are met.