If you discover a potential security issue in this project, please do not create a public GitHub issue. Instead, report it to AWS/Amazon Security via the AWS Vulnerability Reporting page or email aws-security@amazon.com. See CONTRIBUTING.md for details.

PECOS serializes some model and preprocessor artifacts with Python's pickle module (for example, the SklearnTfidf and SklearnHashing vectorizers persist a vectorizer.pkl). Deserializing a pickle file can execute arbitrary code.

Only load model, vectorizer, or preprocessor folders that come from a source you fully trust and control. Do not load artifacts whose path or contents can be influenced by an untrusted party — for example on shared or multi-tenant infrastructure, from shared caches, or from artifact paths derived from external input. This is standard practice for pickle-backed ML artifacts; PECOS is designed for trusted local/batch workflows.