transcribe-anything CVE-2026-79792

Exploitation requires that attacker-controlled input reaches the url argument of the ytdlp_download function without prior sanitization - meaning the application must expose this function to external or untrusted input, such as through a web endpoint, API, or CLI invoked with user-supplied arguments. The CVSS 4.0 vector specifies PR:N and UI:N, meaning no authentication or user interaction is required once the attack path is reachable. However, AC:H indicates high attack complexity - the description characterizes exploitability as 'difficult', suggesting that specific conditions (such as the precise format of the injection payload or timing) must align for reliable exploitation. Exploitation is limited to deployments where the ytdlp_download function accepts externally sourced URLs; local-only or fully automated pipelines fed from trusted sources are not realistically exploitable. Platform and OS are not constrained by the CPE or description.

Despite the CWE-78 classification, which typically signals high severity, the aggregate risk signals are mixed and should be interpreted carefully. The CVSS 4.0 base score is 2.9, driven by AC:H (High Attack Complexity), which the description characterizes as 'difficult' to exploit - suggesting the attacker cannot reliably trigger the injection path without specific preconditions aligning, such as controlling the URL input passed into the function. Impact metrics are uniformly Low (VC:L/VI:L/VA:L) with no subsequent-system impact, constraining the blast radius. On the other hand, a proof-of-concept is publicly available (E:P in the CVSS 4.0 supplemental vector), and PR:N/UI:N means no authentication or user interaction is needed if the attack path is reachable. EPSS data was not provided, so exploitation probability from that dimension is unknown. The vulnerability is not in CISA KEV, and there is no vendor patch. The net assessment is that this represents a moderate-priority issue for any deployment that exposes the ytdlp_download function to attacker-controllable URL input - the public POC lowers the barrier for motivated actors despite the rated high complexity, and the absence of a vendor patch means no quick remediation path exists.

An attacker who can supply a URL argument to the ytdlp_download function - for example, via a web interface or API that passes user input directly into the function - crafts a URL containing embedded shell metacharacters such as ; id # or $(whoami). Because the function does not sanitize the input before constructing the OS command, the injected payload executes in the context of the process running transcribe-anything. A public proof-of-concept demonstrating this technique exists in GitHub issue #137, lowering the skill bar for exploitation once the attack path is reachable.