The vulnerability lies in the download_from_url function of whisperX's FileService. The function makes an HTTP GET request to a user‑provided URL without performing any validation first. Only after the request succeeds does the code check the file extension, and that check can be bypassed by appending .mp3 to the target URL. This flaw permits an attacker to trigger outgoing HTTP requests from the server to arbitrary destinations, exposing internal IP addresses or resources that are otherwise inaccessible from the outside. The flaw is essentially a server‑side request forgery (SSRF) that could be used to harvest sensitive data, scan internal services, or exfiltrate data.

The affected product is the whisperX‑FastAPI application maintained by pavelzbornik. Versions starting at 0.3.1 and up through 0.5.0 contain the vulnerable download_from_url implementation. All earlier releases are unaffected, and the issue was addressed in 0.6.0. Users running any of these vulnerable versions are required to update to 0.6.0 or later to eliminate the backend request vulnerability.

The CVSS base score of 5.8 indicates a medium severity. No EPSS score is reported, and the vulnerability has not been listed in the CISA KEV catalog. Because the exposed /speech‑to‑text‑url endpoint is unauthenticated, an attacker can directly craft malicious requests from outside the network, making the vector local API access with no prior compromise. An internal host that hosts the whisperX service can initiate outbound traffic to any address, including private network segments, thereby bypassing typical network perimeter defenses. The combination of medium CVSS, lack of exploitation data, and open access creates a non‑trivial risk for organizations that deploy the vulnerable versions without additional network controls.