Reported by 4 outlets
CISA Warns of Critical Ray RCE Flaw Exploited in Active Attacks
Why this score
- Breadth
- 2.00 / 2.0 · 4 sources
- Authority
- 1.00 / 2.0 · best outlet weight
- Severity
- 4.60 / 5.0 · CVSS + KEV signals
- Velocity
- 0.00 / 2.0 · updates in 48h
- Pickup speed
- 0.60 / 0.6 · time to second source
- Recency
- × 0.00 temporal multiplier
- Community
- + 0.00 / 1.5 · reddit signal
- Score
- 0.0 / 10 = base × recency + community
Desk analysis
CVE-2025-62593 is a critical RCE, CVSS 4.0 base 9.4, CWE-94 code injection combined with CWE-352 CSRF, in Ray releases before 2.52.0. Ray has never authenticated its Jobs API endpoints /api/jobs and /api/job_agent/jobs/, and its only defense against browser delivery is a check that the User-Agent header starts with "Mozilla". Firefox and Safari allow a page to set that header through fetch, which Chrome refuses because of a Chromium bug, so a DNS rebinding attack against a developer browsing with Firefox or Safari reaches the local Ray dashboard on port 8265 and submits arbitrary code as a job, with no exposed port and no credentials. Anyscale published the advisory GHSA-q279-jhrf-cc6v on November 26, 2025 with a working PoC built on nccgroup/singularity and fix commit 70e7c72780bdec075dba6cad1afe0832772bfe09, and CISA added the CVE to the KEV catalog on August 17, 2026 confirming exploitation in active attacks, with federal remediation due August 20.
The exposed population is developer workstations, not a server fleet, and that changes what defense looks like. The advisory states the same browser flow works as a confused-deputy pivot against network-adjacent Ray instances inside a private corporate network, so one developer loading a malvertised page can hand an attacker code execution in the subnet where training jobs and their cloud credentials live, a path no external attack-surface scanner will ever flag because nothing is exposed to the internet. The KEV confirmation also has history the event stream does not show: Bitsight honeypot telemetry published March 11, 2026 caught the RondoDox botnet using an exploit for CVE-2025-62593 on November 24, 2025, two days before the CVE published, so working exploit code circulated ahead of coordinated disclosure. The same analysis shows RondoDox's implementation defeats itself, because its payload sends User-Agent Mozilla/5.0 (rondo2012@atomicmail.io) and trips the 405 browser guard the exploit exists to bypass, which means the exploitation CISA confirmed most likely involves actors more capable than the one currently visible in honeypots.
Upgrade every Ray install to 2.52.0 or later, then enable the token authentication feature that ships in 2.52.0 disabled by default, because the fix closes the browser path while the Jobs API itself remains unauthenticated, and federal civilian agencies have an August 20, 2026 deadline. For the hunt, review Ray dashboard logs on port 8265 for POSTs to /api/jobs/ or /api/job_agent/jobs/ whose User-Agent does not start with Mozilla, since the public PoC sends the literal value "Other", and for job submissions carrying metadata source "nccgroup/singluarity" or shell entrypoints. HTTP 405 responses reading "Browser requests not allowed" mean someone probed the guard. In DNS telemetry, flag external hostnames resolving to RFC1918 addresses, the rebinding signature, and on endpoints watch for calculator binaries or unexpected shells spawned by ray or dashboard processes on developer machines. Alert on the RondoDox payload hosts rondo.jbt.sh and rondo.zqq.sh and on User-Agents containing rondo2012@atomicmail.io or bang2012@protonmail.com. Where hosts cannot be patched quickly, restrict port 8265 to loopback or a management network and keep browsers off machines that run Ray.
Two developments would move this assessment. A working browser-delivered variant, malvertising or phishing paired with DNS rebinding against Firefox and Safari users, would turn this from botnet shotgun noise into quiet developer-workstation compromise that perimeter telemetry never sees, and today the only publicly documented in-the-wild implementation is self-defeating. Attribution of the KEV exploitation to an actor beyond RondoDox, or telemetry showing mass scanning of directly exposed dashboards on port 8265, would confirm the higher-capability path, and any plain HTTP client that omits a Mozilla User-Agent was never stopped by the guard, so exposed instances need no browser at all.
Timeline
-
update from github.com
Critical RCE Vulnerability against Ray Devs exploitable via Browser (Safari & Firefox) due to DNS Rebinding Attack
github.com · 322d ago · November 26th, 2025
Anyscale's vendor advisory for CVE-2025-62593, published November 26, 2025, describes unauthenticated /api/jobs and /api/job_agent/jobs/ endpoints guarded only by a User-Agent prefix check for "Mozilla", which Firefox and Safari can bypass because their fetch implementation allows setting that header. Combined with DNS rebinding, a developer visiting a malicious page or served malvertising can have arbitrary code submitted to the local Ray dashboard on port 8265, and the same browser flow can pivot at network-adjacent Ray instances inside a private network. Patched in 2.52.0 (fix commit 70e7c72780bdec075dba6cad1afe0832772bfe09), which also introduces a disabled-by-default token authentication feature; the PoC ships in nccgroup/singularity, credit avilum of Oligo as finder and JLLeitschuh of Socket for the PoC and disclosure.
-
update from nvd.nist.gov
NVD Detail: CVE-2025-62593
nvd.nist.gov · 322d ago · November 26th, 2025
NVD published the CVE-2025-62593 record on November 26, 2025 with GitHub, Inc. as CNA: CVSS 4.0 base 9.4 CRITICAL (CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H), affected product ray-project ray below 2.52.0, mapped to CWE-94 and CWE-352. The entry links the GitHub advisory and the patch commit, and records the KEV listing "Ray-Project Ray Code Injection Vulnerability" added August 17, 2026 with a due date of August 20, 2026.
-
update from bitsight.com
RondoDox Botnet: From Zero to 174 Exploited Vulnerabilities
bitsight.com · 217d ago · March 11th, 2026
Bitsight TRACE honeypot research on the RondoDox botnet (174 distinct exploits mapping to 148 CVEs, peaking at 15,000 exploitation attempts in a day) places CVE-2025-62593 in its exploit set with first observed use on November 24, 2025, two days before the CVE published on November 26, 2025, evidence exploit code circulated ahead of coordinated disclosure. The analysis also shows RondoDox's Ray exploit is currently ineffective because its User-Agent "Mozilla/5.0 (rondo2012@atomicmail.io)" triggers the very 405 browser guard the attack must bypass. IoCs include the payload scripts rondo.jbt.sh and rondo.zqq.sh and User-Agents embedding rondo2012@atomicmail.io or bang2012@protonmail.com.
-
update from cyberpress.org
CISA Warns of Critical Ray RCE Flaw Exploited in Active Attacks
cyberpress.org · 56d ago · August 18th, 2026
Cybersecurity and Infrastructure Security Agency (CISA) has added a critical remote code execution vulnerability in Ray, the open-source distributed computing framework, to its Known Exploited Vulnerabilities (KEV) Catalog after confirming exploitation in the wild. Tracked as CVE-2025-62593, the flaw affects Ray versions prior to 2.52.0 and carries a CVSS 4.0 score of 9.4. federal civilian agencies to remediate it by August 20. The agency’s listing states that the vulnerability could allow remote code execution and specifically warns that developers using Ray as a development tool may be exposed when using Firefox or Safari. An attacker can abuse these interfaces to submit or trigger execution of arbitrary code against a vulnerable local Ray instance. The weaknesses are categorized as CWE-94, improper control of code generation (commonly known as code injection), and CWE-352, cross-site request forgery (CSRF) . Code-injection weaknesses occur when externally influenced input can alter the intended syntax or behavior of generated code, potentially enabling unauthorized command execution. According to CISA , Ray’s browser-oriented protection relied on checking whether the User-Agent request header began with “Mozilla.” That safeguard is insufficient