LightLLM version 1.2.0 and earlier do not validate the image_url and audio_url parameters exposed by the multimodal endpoints. This omission permits an unauthenticated attacker to supply arbitrary URLs, resulting in the application fetching internal resources. The retrieved contents or generated error responses can expose sensitive data or reveal internal network topology. The flaw is a classic Server‑Side Request Forgery (CWE‑918).

The vulnerability affects the ModelTC:LightLLM product. All releases up to and including 1.2.0 are impacted. Any deployment of LightLLM 1.2.0 or earlier that exposes the multimodal endpoints is susceptible.

The CVSS score of 6.9 classifies the flaw as medium severity. Because the exploit is allowed without credentials and can reach internal hosts, the risk for organizations that expose the multimodal APIs is significant. The EPSS metric is not available, and the issue is not listed in CISA KEV catalog, but the lack of input validation combined with potential internal network leakage means a careful assessment is warranted. Exploitation would require an attacker to direct the application to a target URL, which can be done by sending a request to the vulnerable endpoint. Attackers could then harvest internal files, URLs, or network information, with only the reach of the application’s outbound network connections as a limiting factor.