The vulnerability resides in the Safetensors Tensor Loader of turboderp ExLlamaV2, specifically within STFile.get_tensor. Manipulating the data_offsets argument triggers a buffer overflow that can compromise memory integrity. If successfully exploited, an attacker could potentially execute arbitrary code or crash the process, resulting in loss of confidentiality, integrity, or availability.

Version 0.3.2 and earlier of turboderp ExLlamaV2 are affected. The flaw is located in the stloader.py component; any deployment using these versions without a patch is vulnerable.

The CVSS score of 5.3 indicates moderate severity. No EPSS data is available and the vulnerability is not listed in the CISA KEV catalog. The attack can be performed remotely by supplying a crafted data_offsets value, and the public exploit demonstrates the feasibility of the attack. As the vendor has not yet responded, attackers may target exposed services that use the vulnerable component.