Disclosed weakness · no CVE recorded

LiteLLM URL-selected models could bypass a key’s model restrictions

LiteLLM · LiteLLM proxy · Google generation and image routes

First documented disclosure · Review checkpoint

Summary

A key restricted to one model could invoke a different model selected by the request URL.

Severity and exploitation evidence

No CVSS score recorded in this review.

Exploitation not established: The reviewed sources describe a correction or controlled reproduction; malicious exploitation against users is not established.

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

Authorization read the allowed model from the body while the route handler dispatched the model in the URL.

Conditions required

An authenticated key holder sends conflicting URL and body model selections to an affected generation or image route.

Potential impact

A model outside the key’s permitted scope could be invoked and billed; conflicting selections could also avoid the intended model budget check.

Affected versions and configurations

The reviewed upstream report does not establish a complete affected-release range.

Fix and mitigation status

1.111.0-rc.1 contains the correction and remains a prerelease. A stable fix was not independently verified.

Practical next steps

Apply the specific correction or evaluate the 1.111.0-rc.1 prerelease under the operator’s release policy; restrict affected routes until corrected.

Operating lesson

Authorize the exact resource selected for execution.

Evidence and limits

The October 28 issue precedes the November 11 correction. The source uses a controlled proxy and deterministic upstream. Tagged verification: the authorization utility uses the shared inference-model resolver and endpoint-selected model semantics. This is distinct from older chat-route reports linked by the issue.

Sources