AI Artifacts
AI artifacts are the digital building blocks, like pre-made models or datasets, that people download to build AI tools. If these files are tampered with by hackers, they can act like a Trojan horse, secretly running harmful commands on your computer the moment you try to use them.
AI artifacts refer to the serialized model files, weights, and training datasets distributed via model registries. These files represent a significant attack vector because they can contain embedded malicious payloads—such as scripts or unsafe serialized objects—that execute arbitrary code when loaded into a production environment.
AI artifacts are the discrete, versioned components of an AI pipeline, including model weights, configuration files, and training corpora, typically managed within centralized registries. Adversaries may achieve full system compromise by introducing malicious AI artifacts, such as models or data, that contain embedded malware or other malicious commands. These artifacts are often stored in model registries or data stores and may affect many systems that pull these resources. Exploitation frequently occurs through the deserialization of insecure formats, such as Python pickle, or the execution of bundled scripts, leading to remote code execution within the host environment.