Federated learning faces threats from malicious clients, with different adversary models impacting the global model accuracy.