Abstract: With the deep integration of cyber domain and physical domain and the fast development of new power systems, cyber attacks pose a severe threat to the safe and reliable operation of power systems, thus it is essential to develop detection methods for cyber attacks. Artificial Intelligence (AI) is recognized as a popular method to detect cyber attacks because of its advantages in extracting data characteristic, modeling complex systems and solving nonlinear systems. This paper first investigates three characteristics of structure complicacy, cyber physical coupling and intelligence in new power systems, and reveals the possible cyber threats that new power systems might encounter in the physical, network and application layers. After that, the AI-based detection methods for cyber attacks in new power systems are reviewed from the perspectives of terminal devices of the physical layer, network layer traffic, packets and business systems in the application layer. Finally, the coupling relationships among attack detection, attack blocking and ex-post recovery are studied, some key technologies of active defense to cyber attacks are discussed, and the corresponding future work are also given.