With artificial intelligence (AI) is extensively applied in education, human-centered AI (HCAI) has become an active field. There although has been increasing concern about how to systematically enhance the AI applications effect, AI risk governance in HCAI education has not been discussed yet. This study adopted literature meta-analysis, along with the Delphi and analytic hierarchy process (AHP) methods in order to establish the risk framework and calculate the index weight of HCAI education. The results confirm that the risk framework includes eight indicators, which respectively are misunderstanding of the HCAI concept (MC), misuse of AI resources (MR), mismatching of AI pedagogy (MP), privacy security risk (PSR), transparency risk (TR), accountability risk (AR), bias risk (BR), and perceived risk (PR). Meanwhile, the eight indicators are divided into four categories such as HCAI concept, application process, ethical security, and man-machine interaction. Moreover, the trend of risks types indicates that more than half of the articles consider only three or less risks types, and the evolution results of risks indicators gradually increased between 2010 and 2021. Additionally, the weights of the eight indicators are MP > MR > AR > PSR > TR > PR > BR > MC. Results obtained could provide theoretical evidence and development suggestions for future scientific governance of HCAI education. Furthermore, the risk framework not only systematically considers the risk governance order of HCAI education, but more importantly, it is the key bridge to the collaborative advancement of stakeholders such as managers, teachers, students, and parents, which can contribute to the scientific, healthy, and sustainable HCAI education.