A patch file loader in HashNeRF-pytorch uses PyTorch’s torch.load to deserialize checkpoint data without validating the input path. According to the description, the manipulation of ckpt_path can trigger deserialization of arbitrary data. It is inferred that a local attacker could potentially execute arbitrary code at the user’s privilege level. This reflects a deserialization flaw (CWE-502) combined with improper input validation (CWE-20), which could compromise confidentiality, integrity, and availability of the system.

The vulnerable code resides in yashbhalgat/HashNeRF-pytorch, specifically the run_nerf.py component prior to commit 82885e698295982504eb6a26d060a6b2473e3706. No discrete version numbers are provided because the project follows a rolling release model, so the exact set of affected releases cannot be enumerated.

With a CVSS score of 4.8 and an EPSS score that is low (under 1%), the overall severity is moderate, but the lack of network exposure limits the attack vector to local users. The vulnerability is not listed in the CISA KEV catalog, suggesting that known exploits are not widespread. Based on the description, it is inferred that a local attacker who loads untrusted checkpoints with torch.load could potentially execute arbitrary code, making this a concern for maintenance and CI environments.