Affected code in the VideoLLaMA2 model loader permits a remote attacker to manipulate the _name_or_path argument and bypass the integrity check that verifies model authenticity. This flaw enables the attacker to load a tampered or malicious model artifact, potentially resulting in arbitrary code execution or corruption of model behavior. The vulnerability is classified as a medium‑severity issue with a CVSS score of 5.3, indicating that it can have significant adverse effects while requiring specific conditions to be fully exploited.

The flaw is present in DAMO Academy VideoLLaMA2 versions up to 1.1, as documented by the vendor. No later versions have been confirmed as affected. The affected component is the model loading module located in videollama2/model/__init__.py.

Exploitability is described as remote and it has been disclosed publicly, suggesting that attackers can trigger the flaw without local access. However, the EPSS score is not available, and the vulnerability is not listed in the CISA KEV catalog, implying that widespread exploitation has not been observed yet. The CVSS of 5.3 places the vulnerability in the medium range, and the lack of a known patch means organizations should treat the risk as moderate but pay close attention to any release notes from DAMO Academy.