The flaw in the server.js API Proxy Endpoint of the openai‑realtime‑ui allows an attacker to manipulate a query argument, causing the server to issue outbound requests to arbitrary URLs. This server‑side request forgery can expose internal resources, pull confidential data, or initiate further attacks by connecting the application to malicious services. The impact includes potential confidentiality and integrity compromise of internal systems, and may also lead to availability issues if the server is overwhelmed.

Any installation of bigsk1's openai‑realtime‑ui that includes the affected component up to commit 188ccde27fdf3d8fab8da81f3893468f53b2797c is vulnerable. Because the project uses continuous delivery with rolling releases, no specific version numbers are published. The flaw remains until the patch commit 54f8f50f43af97c334a881af7b021e84b5b8310f is applied.

The vulnerability carries a CVSS score of 5.3, indicating medium severity. The exploit is remote, publicly available, and has been released, which makes it realistic for adversaries to exploit. The EPSS score is not available and the issue is not listed in the CISA KEV catalog. Exploitation requires only a crafted query to the API Proxy Endpoint, making the attack vector trivial for attackers who can reach the exposed service.