The use of AI to produce ideas and creative works has made artists question the mechanism of AI and the intellectual rights around it. It has been known that AI developers need a set of data to train AI to be able to achieve the desired output. Deepfake, an AI used to produce fake audio recordings, has threatened vocal artists that use their voice as their source of income. Deepfake works by audio sampling the works of actors, singers and voice actors that are available on the internet to produce a fake audio recording of them saying or singing anything desired by the user of the AI program. This paper will analyze the legality of deepfake under Indonesia’s current copyright law number 28/2014 and uncover the practical ways actors, singers and voice actors can defend their works from the use of AI. The method used in this research is the empirical-normative approach because it combines the positive law and the practical side of intellectual rights regarding audio recordings. The research has found that in the fake audio recordings produced by deepfake, copyright cannot be enforced due to the absence of copyrighted work displayed; however if it can be proven that copyrighted works are used as training data for deepfake then it is breaching the rights of performer and copyright owner. In judicial practice however moral rights can’t be sued in court therefore it can only be sued if it is used for commercial purposes and the amount of loss needs to be specified clearly to gain compensation. The solution for this is for the government to make a regulation that enforces deepfake developers to do a license agreement with their imitation target so that the person who owns the voice can be protected from unwanted use of their voice, this also helps developers to obtain training data easily rather than trying to find it on the internet that might be unavailable or illegal. The other option is to force deepfake developers to publish their training data if developers prefer to use existing recordings. Developers that prefer using existing audio recordings must first build a transparent system that ensures the data availability of the sound recordings used on each fake audio recording produced and to manage the legality of each recordings used.