Anyscale Ray is an open-source distributed computing framework that scales Python applications across clusters. It is widely used by data scientists and engineers for machine learning training, data processing, and AI model serving. Ray provides a Dashboard for monitoring cluster health and job status, making it integral to production AI pipelines.
CVE-2023-48023 is a Server-Side Request Forgery (SSRF) vulnerability in the Ray Dashboard. It arises from insufficient input validation on user-supplied URLs, allowing an attacker to trick the server into making unauthorized requests. This flaw is critical because it bypasses network segmentation and exposes internal services.
The vulnerability specifically affects the /log_proxy endpoint of the Ray Dashboard API. When a crafted URL parameter is passed, the server processes it without proper sanitization, enabling requests to internal IPs, cloud metadata endpoints, or other sensitive systems. This endpoint is accessible without authentication in default configurations.
If exploited, an attacker can read internal cloud metadata (e.g., AWS IAM credentials), scan internal networks, or interact with other services behind the firewall. This can lead to data exfiltration, privilege escalation, or full compromise of the Ray cluster and connected infrastructure.
- Upgrade Anyscale Ray to version 2.6.3 or later, which patches the SSRF vulnerability.
- Implement strict input validation on the URL parameter of the /log_proxy endpoint, rejecting internal IP ranges.
- Restrict network access to the Ray Dashboard using firewall rules or security groups to trusted IPs only.
- Enable authentication for the Ray Dashboard to prevent unauthenticated access to sensitive endpoints.
- Monitor logs for unusual outbound requests from the Ray server, especially to internal IPs or cloud metadata services.
- Apply network segmentation to isolate the Ray cluster from critical internal systems.
- Use a web application firewall (WAF) to filter malicious payloads targeting SSRF vulnerabilities.
- Conduct regular security assessments and penetration testing on Ray deployments.
Get AI-powered remediation steps tailored to your asset.
Try AI Solutions →