GHSA-xp4f-hrf8-rxw7: picklescan: scanner bypass leads to undetected RCE

### Summary Using ensurepip._run_pip 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 ensurepip._run_pip 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 ensurepip import _run_pip class EvilEnsurepipRunpip: def __reduce__(self): payload = "[(__import__('os').system('whoami'),)]" return _run_pip, (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 targeting a financial institution's fraud-detection ML pipeline crafts a malicious PyTorch model using the ensurepip._run_pip reduce gadget. The payload establishes a reverse shell or exfiltrates environment variables containing API keys. The attacker publishes this model to a public Hugging Face repository or submits it via a vendor. The victim's MLOps pipeline runs picklescan (< 0.0.30) as the security gate — it returns clean. The pipeline promotes the model to staging, where pickle.load() triggers RCE. The attacker now has code execution inside the ML inference environment, which typically has broad internal network access and cloud IAM credentials attached.