A weakness has been identified in ModelTC LightX2V up to 0.5.0 that allows manipulation of the auto_map argument within the AutoTokenizer.from_pretrained function. The manipulation can lead to code injection, granting an attacker the ability to execute arbitrary code within the local environment where the tokenizer is instantiated. This is a local-only vulnerability that can compromise confidentiality, integrity, and availability of the affected system. The defect is consistent with CWE-74 (Unexpected URL Redirects) and CWE-94 (Improper Control of Generation of Code).

The vulnerability affects the ModelTC LightX2V product, specifically versions up to and including 0.5.0. The issue resides in the file lightx2v/models/runners/cosmos3/cosmos3_runner.py within the Tokenizer component. No other product or version information is provided.

The CVSS score of 4.8 indicates a moderate security impact. EPSS data is not available, and the vulnerability is not listed in the CISA KEV catalog. The attack is limited to local execution, meaning the attacker must have access to the host running the application or have the ability to run code locally. Because the exploit was made publicly available, a threat actor who can supply a crafted auto_map value can trigger code injection if the vulnerable function is called in a context that permits such input. Consequently, the risk is primarily to systems where the tokenizer is used without proper input validation or where the application runs with elevated privileges.