In a model file attack, attackers upload malicious AI model files to trusted open source repositories. These files appear legitimate but contain hidden executable code that executes automatically when a developer loads the model, potentially stealing credentials and exfiltrating data.