Critical RCE in Ray AI Compute Engine (CVE‑2025‑62593) Added to CISA KEV Catalog
What It Is — Ray 2.52.0 is an open‑source AI compute engine that exposes a dashboard/API. Versions < 2.52.0 relied on a naïve check of the HTTP User‑Agent header (“Mozilla”) to block browser‑based attacks. This check can be bypassed, allowing a DNS‑rebinding attack to deliver arbitrary code to a developer’s browser session.
Exploitability — The vulnerability is actively exploited in the wild; CISA has placed it in the Known Exploited Vulnerabilities (KEV) catalog. A public proof‑of‑concept exists that combines header spoofing with DNS rebinding. CVSS v3.1 = 9.4 (Critical).
Affected Products — Ray AI Compute Engine (all releases prior to 2.52.0). The flaw impacts any environment where the dashboard is reachable from a web browser (e.g., Firefox, Safari).
Why It Matters for Compliance & Audit Readiness
- Control Mapping – The issue maps to SOC 2 CC6.1 (system operations) and CC7.2 (change management); unpatched systems constitute a control gap that auditors will flag.
- Continuous Evidence – Demonstrating timely patch deployment and verification of dashboard hardening provides concrete evidence for audit trails and for meeting CISA’s BOD 22‑01 deadline.
- Enterprise Buyer Expectations – Federal agencies and large enterprises now require proof that known‑exploited flaws are remediated within mandated windows; a documented remediation workflow satisfies that demand.
Recommended Actions
- Upgrade all Ray installations to 2.52.0 or later immediately.
- Verify that the dashboard/API is restricted to trusted networks or protected by strong authentication (e.g., MFA, zero‑trust gateway).
- Add the CVE to your vulnerability‑management inventory and map it to the relevant SOC 2 controls; capture patch‑status screenshots as audit evidence.
- Implement continuous monitoring for DNS‑rebinding attempts and anomalous User‑Agent strings.
- Review CISA’s Binding Operational Directive 22‑01 and ensure remediation by the agency‑wide deadline (Aug 20 2026).
Source: SecurityAffairs article