A vulnerability in PyTorch's weights_only unpickler allows remote attackers to execute arbitrary code by loading a malicious checkpoint (.pth) file. The flaw involves improper validation of pickle opcodes and storage metadata, leading to heap memory corruption and storage size mismatches. This can be exploited if a user loads a crafted checkpoint file with torch.load(..., weights_only=True). The issue affects PyTorch versions prior to 2.10.0 and is fixed in version 2.10.0.