Disclosed weakness · no CVE recorded

SurrealDB model exports can send session credentials to another host

SurrealDB · SurrealDB remote HTTP SDK machine-learning export

First documented disclosure · Review checkpoint

Summary

A model name interpreted as a URL can redirect an authenticated model-export request to another host.

Severity and exploitation evidence

High · numeric CVSS score not recorded. Severity source.

Exploitation not established: The maintainer verified the redirect mechanism. Malicious exploitation and credential reuse are unestablished.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

The remote SDK joins an untrusted model name onto a URL. An absolute URL replaces the configured origin, and the client attaches its existing session headers to the new destination.

Conditions required

An authenticated SDK client exports a model using a name the attacker can influence, and the client can reach the chosen host.

Potential impact

The destination can receive session credentials and authenticated requests. Subsequent access depends on the leaked credential’s permissions.

Affected versions and configurations

Remote HTTP clients with the affected model export path before 4.0.0, per SurrealDB.

Fix and mitigation status

4.0.0 pins the export destination to its configured origin.

Practical next steps

Update the client SDK. Review model exports to unexpected hosts and revoke credentials if an export reached an untrusted destination.

Operating lesson

Verify the final host before attaching authentication to a request constructed from a model name.

Evidence and limits

November 3 is the documented public release and advisory date. An October 21 commit timestamp alone does not establish public disclosure. The maintainer rates the flaw High without a numeric CVSS or CVE.

Sources