The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has added CVE-2025-62593 Ray to its Known Exploited Vulnerabilities (KEV) catalog. The reason: CISA cites evidence that the vulnerability is being actively exploited in real-world scenarios.
The issue affects Ray, an open-source, Python-native distributed computing framework used to scale machine learning and AI workloads. In practice, it can be leveraged through a browser-based path to trigger remote code execution—a serious escalation for developers and organizations that run Ray in development and testing environments.
What CVE-2025-62593 Ray enables
The vulnerability, tracked as CVE-2025-62593, carries a CVSS score of 9.4. CISA’s KEV designation is based on evidence that attackers are using it, but the public details about the exact exploitation method are limited.
According to the Ray maintainers’ advisory from 2025, the flaw allows an attacker to execute arbitrary code against Ray by targeting critical endpoints that do not enforce authentication. The browser is a central part of the attack chain.
DNS rebinding and the browser-based RCE risk
At the core is a browser attack that can be made possible through a DNS rebinding technique. Ray maintainers specifically describe the attack in terms of a modified User-Agent header combined with DNS rebinding against the browser.
That combination enables remote code execution via web browsers such as Mozilla Firefox and Apple Safari. In other words, the problem is not limited to a direct network connection to Ray; it can be triggered when a developer’s machine interacts with malicious web content.
Who is most at risk
The maintainers note that the defect primarily impacts developers running development or testing environments with Ray. This matters because those setups are often less strictly hardened than production systems and may include exposed services or default configurations.
If a targeted developer falls for a phishing attempt, or visits a site that serves a malicious advertisement, the attacker may be able to reach the vulnerable Ray functionality from within the browser. The result can be execution of arbitrary shell code on the developer’s machine.
There is also an additional pathway: the maintainers explain that the approach can be extended to affect network-adjacent Ray instances. In that case, the browser can act as a confused deputy intermediary, helping shift the impact toward Ray deployments reachable from within a private corporate network.
What CISA knows—and what it doesn’t
CISA has not publicly released technical details about the exploitation mechanics “in the wild.” However, other reporting points to how quickly the vulnerability was adopted once proof-of-concept material became available.
For example, a BitSight report from March 2026 indicated that attackers associated with the RondoDox DDoS botnet incorporated the vulnerability into their toolkit two days before it was publicly disclosed on November 26, 2025. The report attributes this rapid adoption to the availability of a proof-of-concept exploit.
Related campaigns: from botnets to GPU mining
The Ray maintainer community also links unpatched Ray instances to other malicious activity. Oligo reported that attackers aimed to repurpose compromised clusters equipped with NVIDIA GPUs into a self-replicating cryptocurrency mining botnet.
This activity was described under the campaign name ShadowRay 2.0. While this does not replace the need to address the browser-based RCE threat, it reinforces the broader point: running Ray without timely fixes can create an entry point for multiple threat behaviors.
How to mitigate: update Ray
The good news is that the issue is addressed in a specific release. Ray maintainers state that the vulnerability has been fixed in version 2.52.0 of the Python package.
If you run Ray in any environment where a developer workstation could access Ray endpoints—especially when testing or using local development setups—patching should be treated as urgent. In addition to updating, organizations should review exposure patterns and endpoint protections to reduce the impact of browser-triggered attacks.
Ray also credited Oligo security researcher Avi Lumelsky for discovering a fetch bypass, and Jonathan Leitschuh for developing the DNS rebinding attack concept. That context suggests the vulnerability hinged on subtle logic and request handling weaknesses that were not previously closed by design.
CISA’s deadline for federal agencies
Because CVE-2025-625-62593 Ray is in active exploitation territory, CISA recommends that Federal Civilian Executive Branch (FCEB) agencies apply the required fixes and mitigations by August 20, 2026.
Even if you are outside the federal scope, the KEV listing should be treated as an indicator that attackers are already moving beyond theory. Waiting for your next maintenance window can increase exposure, particularly for teams that commonly browse untrusted content during daily work.
Practical steps you can take now
- Upgrade Ray to 2.52.0 or later, and verify the version across development and testing machines.
- Audit where Ray endpoints are reachable from developer workstations and internal networks.
- Harden configurations and apply strong access controls so that sensitive Ray endpoints cannot be triggered without authorization.
- Monitor for suspicious browser-driven activity, especially around Ray endpoints used by developer workflows.
- Plan for remediation beyond patching by reducing the trust and exposure level of development environments.
Bottom line
CISA’s KEV update for CVE-2025-62593 Ray is a clear signal that this is more than a potential issue. The vulnerability enables browser-based remote code execution by combining insufficient endpoint protections with a DNS rebinding attack chain—meaning attackers can target developers through malicious web interactions.
Update to Ray 2.52.0 and review how Ray endpoints are exposed in development and testing environments. Doing so quickly reduces the likelihood that a user’s browsing session becomes the trigger for arbitrary code execution.
Source: https://thehackernews.com/2026/08/cisa-flags-actively-exploited-ray-flaw.html
