The digital transformation of government has led to the adoption of artificial intelligence (AI) to support public services, including complaint chatbots, automated social assistance eligibility assessment systems, and algorithm-based licensing. However, under administrative law, public decisions must be issued by authorized officials in accordance with the principles of legality and accountability. When part of the decision-making process is conducted through an automated system, important legal questions arise: Can automated decisions be categorized as a form of administrative discretion, and who bears legal responsibility when such decisions harm citizens? This study aims to analyze the legal status of AI-based automated decisions in relation to the concept of administrative discretion and to formulate a legal framework that ensures accountability, transparency, and the protection of citizens’ rights. This research employs a normative juridical method using statutory, conceptual, and comparative approaches. The analysis focuses on Articles 22–23 of Law No. 30 of 2014 concerning Government Administration, relevant regulations on electronic-based government systems, and practices of digital public services. The findings show that automated decisions cannot be regarded as administrative discretion without a clear legal basis authorizing their use, while legal responsibility remains with the authorized officials or public institutions operating the system. The novelty of this study lies in its conceptualization of “digital discretion” within Indonesian administrative law and its proposal of a risk-based supervision model incorporating a human-in-the-loop mechanism to ensure that the use of AI in public administration remains subject to the rule of law.