CVE-2026-34981: WhisperX API Auth Bypass Vulnerability

CVE-2026-34981 Overview

CVE-2026-34981 is a Server-Side Request Forgery (SSRF) vulnerability affecting the whisperX-FastAPI, a tool for enhancing and analyzing audio content. The vulnerability exists in the FileService.download_from_url() function within app/services/file_service.py, which makes HTTP requests without proper URL validation. An attacker can exploit this flaw to access internal network resources and potentially exfiltrate sensitive information from systems that should not be accessible from the internet.

Critical Impact

Unauthenticated attackers can make arbitrary HTTP requests to internal network resources through the /speech-to-text-url endpoint, potentially accessing cloud metadata services, internal APIs, and other protected resources.

Affected Products

• whisperX-FastAPI versions 0.3.1 through 0.5.0

Discovery Timeline

• 2026-04-06 - CVE CVE-2026-34981 published to NVD
• 2026-04-07 - Last updated in NVD database

Technical Details for CVE-2026-34981

Vulnerability Analysis

This Server-Side Request Forgery vulnerability stems from a fundamental flaw in input validation within the whisperX-FastAPI application. The FileService.download_from_url() function directly calls requests.get(url) on user-supplied URLs without first validating whether the URL points to an allowed external resource or a potentially dangerous internal network location.

The vulnerability is classified under CWE-918 (Server-Side Request Forgery), which describes scenarios where a web application fetches remote resources without properly validating user-supplied URLs. This allows attackers to coerce the server into making requests to arbitrary destinations.

A critical aspect of this vulnerability is that while the application does implement a file extension check, this validation occurs after the HTTP request has already been made. This ordering mistake means the server will connect to any specified URL before any security checks are applied. Furthermore, the extension validation can be trivially bypassed by appending .mp3 to any internal URL, allowing attackers to target virtually any internal resource while appearing to request an audio file.

Root Cause

The root cause is the lack of URL validation before making outbound HTTP requests in the FileService.download_from_url() function. The code directly passes user-controlled URL input to requests.get() without checking whether the target is an internal IP address, localhost, cloud metadata endpoint, or other restricted destination. The post-request file extension check provides no meaningful security as the damage (the unauthorized request) has already occurred by the time validation happens.

Attack Vector

The attack leverages the unauthenticated /speech-to-text-url endpoint to submit malicious URLs. An attacker crafts a URL pointing to internal network resources (such as http://169.254.169.254/latest/meta-data/.mp3 for AWS metadata or http://localhost:8080/admin/.mp3 for internal services), and the server dutifully fetches the content on behalf of the attacker. The .mp3 suffix bypass allows targeting any internal endpoint regardless of the actual content type, enabling reconnaissance of internal networks, access to cloud instance metadata, and potential retrieval of sensitive credentials or configuration data.

Detection Methods for CVE-2026-34981

Indicators of Compromise

• Unexpected outbound HTTP requests from the whisperX-FastAPI server to internal IP ranges (10.x.x.x, 172.16.x.x, 192.168.x.x)
• Access logs showing requests to /speech-to-text-url with URLs targeting cloud metadata endpoints (169.254.169.254)
• Unusual network connections from the application server to localhost or link-local addresses
• HTTP requests containing suspicious URL patterns with .mp3 appended to non-audio resource paths

Detection Strategies

• Configure web application firewalls (WAF) to inspect URL parameters for internal IP addresses and metadata service endpoints
• Implement network monitoring to detect and alert on outbound connections from the whisperX-FastAPI service to internal network ranges
• Enable detailed application logging for all requests to the /speech-to-text-url endpoint and review for anomalous URL patterns
• Deploy intrusion detection rules to identify SSRF attack patterns in HTTP request parameters

Monitoring Recommendations

• Monitor egress traffic from application servers for connections to RFC 1918 private IP ranges and cloud metadata services
• Set up alerts for high volumes of requests to the /speech-to-text-url endpoint from single source IPs
• Review application logs for URL parameters containing localhost, internal hostnames, or IP addresses with non-standard ports
• Implement real-time analysis of outbound HTTP request destinations from the application tier

How to Mitigate CVE-2026-34981

Immediate Actions Required

• Upgrade whisperX-FastAPI to version 0.6.0 or later immediately
• If immediate upgrade is not possible, restrict network access to the /speech-to-text-url endpoint using firewall rules or reverse proxy configuration
• Implement network-level egress filtering to prevent the application from connecting to internal IP ranges
• Consider temporarily disabling the /speech-to-text-url endpoint until patching is complete

Patch Information

The vulnerability has been fixed in whisperX-FastAPI version 0.6.0. The security fix is documented in GitHub Security Advisory GHSA-6rc7-r867-c635 and implemented in commit ef78fe2. Additional details about the vulnerability are available in GitHub Issue #256.

Workarounds

• Deploy a reverse proxy in front of the application that filters requests to the vulnerable endpoint based on URL parameter inspection
• Implement network segmentation to prevent the whisperX-FastAPI server from reaching internal resources
• Add authentication requirements to the /speech-to-text-url endpoint using middleware or API gateway policies
• Use egress firewall rules to restrict the application server's outbound connectivity to only necessary external destinations

# Example: Block metadata endpoint access using iptables
iptables -A OUTPUT -d 169.254.169.254 -j DROP
iptables -A OUTPUT -d 10.0.0.0/8 -j DROP
iptables -A OUTPUT -d 172.16.0.0/12 -j DROP
iptables -A OUTPUT -d 192.168.0.0/16 -j DROP