GHSA-m869-42cg-3xwr: picklescan: scanner bypass enables RCE via ML models

### Summary Using idlelib.run.Executive.runcode function, 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.run.Executive.runcode 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 ``` from idlelib.run import Executive from types import SimpleNamespace class EvilIdlelibRunExecutiveRuncode: def __reduce__(self): payload = "__import__('os').system('whoami')" fake_self = SimpleNamespace(locals={}) return Executive.runcode, (fake_self, payload) ``` ### 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 adversary targets an organization that pulls open-source models from HuggingFace for fine-tuning. They publish a malicious PyTorch model (.pth) embedding the exploit via the EvilIdlelibRunExecutiveRuncode payload — a Python class whose __reduce__ method invokes idlelib.run.Executive.runcode with an arbitrary command string. The organization's MLOps pipeline runs picklescan on all downloaded models; it reports clean. A data scientist runs torch.load() on the model — this deserializes the pickle, triggers __reduce__, and executes the attacker's payload (e.g., reverse shell or credential exfiltration) on the ML training server, which typically has broad AWS/GCP IAM permissions for dataset access.