CVE-2025-62593
None CISA KEVCVSS
—
EPSS
62.5%
Published
Nov 26, 2025
CISA due date
Aug 20, 2026
Cloudanix analysis
A flaw in the Ray AI tool lets attackers take over a developer's computer if the developer simply visits a bad website while the tool is running.
- Exploitation
- Requires User Interaction
- Attacker needs
- User Interaction
- Complexity
- Moderate
- Impact
- Remote Code Execution
- Component
- Application
- Class
- RCE
- Remediation
- Upgrade
- Effort
- Medium
- Patch risk
- Low
What it actually does
Remote code execution on Ray instances via browser-based DNS rebinding attack that bypasses an insufficient User-Agent header check requiring only a leading Mozilla string.
Why it matters to the business
Developers and organizations using Ray face risk of full system compromise from visiting malicious sites or ads, with CISA-mandated remediation required by 2026-08-20 and rapidly rising exploit likelihood.
How an attack unfolds
An attacker hosts a malicious site or malvertisement that triggers a DNS rebinding attack against a browser; the browser then issues requests to the local Ray instance with a spoofed User-Agent, bypassing the weak Mozilla check and achieving code execution. Active exploitation is occurring in the wild by actors targeting the Ray-Project Ray Code Injection Vulnerability.
How to fix it
Upgrade Ray to version 2.52.0 or later. For Debian 13 systems, upgrade to linux package version 6.11.2-1.
How to confirm the fix worked
Confirm Ray version is 2.52.0 or higher and that Debian linux package is at 6.11.2-1; test that non-Mozilla User-Agent strings are rejected for local API access.
Typical attack path
- Victim runs Ray and visits attacker-controlled website or receives malvertising
- Attacker performs DNS rebinding to make browser treat attacker domain as local Ray host
- Browser sends request with attacker-controlled User-Agent header
- Ray accepts request due to weak Mozilla prefix check and executes attacker-supplied code
How to detect it
Review web proxy, browser, and local service logs for unexpected connections to Ray ports from external domains combined with User-Agent strings that start with Mozilla but originate from rebinding attempts; look for process execution or file creation artifacts tied to the Ray-Project Ray Code Injection Vulnerability.
- User-Agent header containing Mozilla prefix followed by unexpected content in requests to Ray ports
- DNS responses showing rapid A-record changes targeting localhost or 127.0.0.1
- Unexpected child processes spawned by ray process after browser activity
What your security team should do
- Immediately upgrade all Ray instances to 2.52.0 and Debian linux packages to 6.11.2-1 before the 2026-08-20 KEV deadline
- Monitor EPSS trend and block outbound DNS rebinding attempts from developer workstations
- Add network controls to prevent browser-initiated connections to local Ray ports from untrusted sites
The record
Ray is an AI compute engine. Prior to version 2.52.0, developers working with Ray as a development tool can be exploited via a critical RCE vulnerability exploitable via Firefox and Safari. This vulnerability is due to an insufficient guard against browser-based attacks, as the current defense uses the User-Agent header starting with the string "Mozilla" as a defense mechanism. This defense is insufficient as the fetch specification allows the User-Agent header to be modified. Combined with a DNS rebinding attack against the browser, and this vulnerability is exploitable against a developer running Ray who inadvertently visits a malicious website, or is served a malicious advertisement (malvertising). This issue has been patched in version 2.52.0.
Affected packages
| Ecosystem | Package | Introduced | Fixed |
|---|---|---|---|
| PyPI | ray | 0 | 2.52.0 |
| Chainguard | airflow-3-compat | 0 | 3.1.3-r0 |
| Chainguard | airflow-3-iamguarded-compat | 0 | 3.1.3-r0 |
| Chainguard | py3-vllm-cuda-12.4 | 0 | 0.11.2-r1 |
| Chainguard | py3.10-vllm-cuda-12.4 | 0 | 0.11.2-r1 |
| Chainguard | py3.12-vllm-cuda-12.4 | 0 | 0.11.2-r1 |
| Chainguard | tritonserver-backend-vllm-cuda-12.9 | 0 | 25.9.0_git20251112-r1 |
| Chainguard | tritonserver-backend-vllm-meta-cuda-12.9 | 0 | 25.9.0_git20251112-r1 |
| Wolfi | airflow-3-compat | 0 | 3.1.3-r0 |
| Wolfi | airflow-3-iamguarded-compat | 0 | 3.1.3-r0 |
Related vulnerabilities
- CVE-2026-27482 Ray is an AI compute engine. Also affects ray
- GHSA-q5fh-2hc8-f6rq Ray dashboard DELETE endpoints allow unauthenticated browser-triggered DoS (Serve… Also affects ray
- CVE-2025-34351 Ray's New Token Authentication is Disabled By Default Also affects ray
- CVE-2023-6020 LFI in Ray's /static/ directory allows attackers to read any file on the server without… Also affects ray
- CVE-2023-6021 LFI in Ray's log API endpoint allows attackers to read any file on the server without… Also affects ray
- CVE-2023-6019 A command injection existed in Ray's cpu_profile URL parameter allowing attackers to… Also affects ray
- CVE-2025-69229 AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. Also affects airflow-3-compat
- CVE-2025-69223 AIOHTTP's HTTP Parser auto_decompress feature is vulnerable to zip bomb Also affects airflow-3-compat