The GIF decoder in PyTorch torchvision, up to version 0.28.0, performs a memcpy using a length that is not properly clamped, creating an out‑of‑bounds heap read. An attacker who can supply a malicious or truncated GIF file can trigger a segmentation fault, causing a denial of service, or can expose adjacent heap memory, potentially leaking sensitive data. The weakness corresponds to CWE‑125.

PyTorch torchvision versions up to and including 0.28.0 are affected. The vulnerability exists in the vision library maintained by the PyTorch project.

The CVSS score is 7.1, indicating moderate to high severity. The EPSS score is less than 1 %, showing a low probability of exploitation in the wild. The vulnerability is not listed in the CISA KEV catalog. Although the description does not specify a remote exploitation method, the likely attack scenario involves an application that processes user‑supplied GIF images; an attacker can supply a crafted GIF to trigger the flaw. The exception is that only code paths that invoke the GIF decoder are relevant, and no exploitation depends on elevated privileges or system configuration.