A model extraction attack is a type of cybersecurity threat targeting machine learning systems, especially those offered as APIs or online services. In this attack, an adversary sends numerous queries to the target AI model, carefully analyzing the outputs to infer the model's parameters, architecture, or training data. By systematically querying the model and collecting responses, attackers can reconstruct a close approximation or even an exact copy of the original model. This can lead to intellectual property theft, loss of competitive advantage, and exposure of sensitive or proprietary information embedded in the model. Model extraction attacks can also facilitate further attacks, such as adversarial examples or model inversion.