PyTorch torchvision through version 0.28.0 contains an out-of-bounds heap read vulnerability in the GIF decoder component. Specifically, the read_from_tensor callback passes an unclamped length parameter to memcpy, which can be exploited by attackers providing crafted or truncated GIF files. This can cause a segmentation fault resulting in denial of service or potentially leak adjacent heap memory contents. The vulnerability is addressed in commit 4e05dc2. No official remediation level or patch link is provided in the available data.

Successful exploitation can cause application crashes (denial of service) due to segmentation faults or disclose adjacent heap memory contents, potentially leaking sensitive information. The vulnerability is exploitable without privileges or user interaction other than supplying a malicious GIF file. No known exploits in the wild have been reported.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. The vulnerability is fixed in commit 4e05dc2; users should update torchvision to a version including this fix once available. Until then, avoid processing untrusted GIF files with affected versions.