GHSA-4whj-rm5r-c2v8: picklescan: scanner bypass enables PyTorch gadget RCE

### Summary Using torch.utils.bottleneck.\_\_main\_\_.run_autograd_prof function, which is a pytorch library function to execute remote pickle file. ### Details The attack payload executes in the following steps: First, the attacker craft the payload by calling to torch.utils.bottleneck.__main__.run_autograd_prof function in reduce method Then when the victim after checking whether the pickle file is safe by using Picklescan library and this library doesn't dectect any dangerous functions, decide to pickle.load() this malicious pickle file, thus lead to remote code execution. ### PoC ``` import torch.utils.bottleneck.__main__ as bottleneck_main class EvilTorchUtilsBottleneckRunAutogradProf: def __reduce__(self): code = '__import__("os").system("whoami")' globs = {} return bottleneck_main.run_autograd_prof, (code, globs) ``` ### Impact Who is impacted? Any organization or individual relying on picklescan to detect malicious pickle files inside PyTorch models. What is the impact? Attackers can embed malicious code in pickle file that remains undetected but executes when the pickle file is loaded. Supply Chain Attack: Attackers can distribute infected pickle files across ML models, APIs, or saved Python objects. ### Corresponding https://github.com/FredericDT https://github.com/Qhaoduoyu

An attacker uploads a malicious PyTorch model to a public registry (HuggingFace, GitHub) or injects it via a compromised internal artifact store. The victim organization's CI/CD pipeline invokes picklescan < 0.0.30, which reports the file as clean. The model is promoted to staging and loaded with torch.load(), triggering the __reduce__ gadget chain through run_autograd_prof, which evaluates arbitrary Python expressions. The attacker achieves RCE on the model server—potentially establishing reverse shell persistence, exfiltrating training data and model IP, or pivoting laterally through the ML infrastructure.