This paper presents a comprehensive survey of AI-generated text detection methods specifically designed for Arabic language content. We analyze current approaches including transformer-based models like AraELECTRA and XLM-R, watermarking techniques, and statistical detection methods. Our analysis reveals that most Arabic-specific methods rely heavily on BERT-based architectures trained on limited datasets, while more sophisticated approaches like synthetic ID watermarking remain underutilized. We identify key limitations in existing benchmarks, particularly the reliance on known poems and weak evaluation metrics. The paper discusses the need for stronger evaluation frameworks, larger diverse datasets, and the integration of advanced detection techniques beyond traditional classification approaches. We provide critical remarks on methodological gaps and propose directions for future research in Arabic AI text detection.