GHSA-7cq8-mj8x-j263: picklescan: detection bypass allows malicious pickle RCE

### Summary Using idlelib.autocomplete.AutoComplete.fetch_completions, which is a built-in python 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 idlelib.autocomplete.AutoComplete.fetch_completions 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 ``` class EvilIdlelibAutocompleteFetchCompletions: def __reduce__(self): from idlelib.autocomplete import AutoComplete, ATTRS return AutoComplete().fetch_completions, ("__import__('os').system('whoami')", ATTRS) ``` ### 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

Adversary uploads a malicious PyTorch model to a public registry (HuggingFace, GitHub, internal artifact store). The model's pickle file encodes a `__reduce__` method that calls `idlelib.autocomplete.AutoComplete().fetch_completions` with an OS command as argument — a Python stdlib function that picklescan does not flag. The target organization's MLOps pipeline runs picklescan as a pre-load validation step; the scan returns clean. The model is loaded into a training or inference worker, triggering RCE. The attacker gains code execution in the ML infrastructure — typically a privileged environment with GPU access, cloud credentials, and access to training data.