[EN] AI Security · Dec 24, 2025 Beyond the Tensors: Exploring the Universal Vulnerabilities of ML Model Formats