Description

nodejs-whisper through 0.3.1 contains an OS command injection vulnerability in convertToWavType() that allows attackers to execute commands by supplying crafted audio file paths. Attackers can provide file names containing $() command substitution or closing quotes, which are interpolated into an ffmpeg command run via shelljs exec(), executing OS commands with Node.js process privileges.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "9.8",
  "pubDate": "2026-10-14T15:17:17.940Z",
  "pubdate": "2026-10-14T15:17:17.940Z",
  "executiveSummary": "The nodejs-whisper package, specifically in versions up to and including 0.3.1, is susceptible to an OS command injection vulnerability located within the convertToWavType() function. This vulnerability arises from improper handling of user-supplied audio file paths, which are directly concatenated into a shell command string.\nBy crafting malicious file paths containing shell metacharacters such as command substitution operators ($()) or closing quotes, an unauthenticated attacker can force the underlying system to execute arbitrary OS commands. The command execution is performed via the shelljs exec() method, meaning the injected payloads inherit the security context and privileges of the Node.js process executing the library.\nThe impact of this vulnerability is severe, as successful exploitation enables remote code execution (RCE) on the host environment, potentially leading to unauthorized data access, system compromise, or complete service disruption. The flaw does not require authentication, posing a significant risk to any application that processes untrusted file inputs through this function. Remediation requires sanitizing input paths or moving away from shell-based command execution interfaces.",
  "technicalDetails": "The vulnerability originates in the convertToWavType() function of the nodejs-whisper library. This component is responsible for preparing audio files for processing by the Whisper engine, which involves invoking the ffmpeg binary via the shelljs exec() function. The root cause of the vulnerability is an unsafe string concatenation pattern where the input parameter representing the audio file path is inserted directly into the shell command string without adequate sanitization or escaping of shell metacharacters.\nExploitation is achieved by manipulating the file path string supplied to the vulnerable function. Because the input is treated as a literal part of the shell command, an attacker can append malicious shell syntax. For instance, by providing a filename such as 'test.mp3; $(malicious_command)', the attacker can terminate the intended ffmpeg command and execute an arbitrary command of their choosing.\nThe attack flow proceeds as follows: First, the attacker provides a maliciously crafted string as the filename input to the convertToWavType() function. Second, the library constructs the final shell string by joining the input path with the necessary ffmpeg flags. Third, this concatenated string is passed to shelljs exec(), which executes the entire command within the context of the default system shell (/bin/sh or cmd.exe). Because shelljs exec() spawns a shell process to interpret the provided string, all contained metacharacters—such as semicolons, backticks, or command substitution constructs like $(...)—are parsed and executed.\nThe vulnerability is present in versions of nodejs-whisper up to 0.3.1. There are no authentication or privilege requirements to trigger this vulnerability, as it is exploitable as long as the application allows processing of attacker-controlled file paths. The exposure depends entirely on the implementation of the host application; if a web service accepts user-uploaded files or file paths and passes them through this function, an attacker can gain arbitrary command execution with the privileges of the Node.js process. Post-exploitation, an attacker can pivot within the local network, exfiltrate sensitive environment variables, or install persistent backdoors on the host server."
}