Impact
A code injection flaw exists in the _load_text_config function of TinyLLaVA TinyLLaVA_Factory, where an attacker can manipulate the llm_model_name_or_path argument to execute arbitrary code. The vulnerability aligns with CWE-74 (Command Injection) and CWE-94 (Dynamic Code Execution). Remote exploitation is possible, and an exploit has already been released to the public. The flaw can allow attackers to run arbitrary commands on the host system or gain full control of the application.
Affected Systems
The flaw affects all releases of TinyLLaVA TinyLLaVA_Factory up to the commit 1abc6ac5e76a68d0fd087a071b56da968e69ca9b. Because the project uses a rolling release model, the exact version numbers for the vulnerable and patched releases are not documented, but any codebase that includes the unpatched _load_text_config function is vulnerable.
Risk and Exploitability
The CVSS score of 5.3 indicates moderate severity. The EPSS score is not available, so exploitation probability is unknown, but the vulnerability is publicly known and no Mitigations are listed in the CISA KEV catalog. Remote exploitation is feasible by sending a crafted llm_model_name_or_path value, allowing an attacker to inject shell commands or Python code into the configuration loader. Given the public release of the exploit, the risk to systems running this software is significant until a fix is applied.
OpenCVE Enrichment