lmdeploy API Server Vulnerability: Potential SSRF Threat
lmdeploy's OpenAI-compatible API server reportedly contains a server-side request forgery vulnerability that could allow unauthenticated attackers to access internal services and cloud metadata endpoints.
What happened
An early warning has been issued regarding a potential critical vulnerability in lmdeploy's API server. The vulnerability, tracked as CVE-2026-63764, appears to allow unauthenticated attackers to perform server-side request forgery (SSRF) attacks by supplying a crafted image_url that redirects to internal targets. This could enable attackers to access sensitive internal services and cloud metadata endpoints.
The vulnerability is under investigation and has been assigned a CVSS score of 9.3, indicating a critical severity level. The attack vector involves sending a POST request to the chat completions endpoint with an image_url pointing to an attacker-controlled server that responds with an HTTP 302 redirect to internal addresses, bypassing initial URL safety checks.
Professional software engineers using lmdeploy are advised to stay informed about updates from primary sources and consider reviewing their internal services for any signs of unauthorized access. It is recommended to upgrade to a patched version of lmdeploy as soon as it becomes available. For the latest information, consult the NVD page for CVE-2026-63764.
How 0Day mitigates this
lmdeploy is anywhere in your dependency tree, the engineers who own the affected repositories get a push alert the moment it is flagged — no manual audit to remember to run.Read how this differs from waiting on a scanner to catch a known advisory, or see the exact, read-only access 0Day needs to do this for an organization.