Thanks to the vigorous development of artificial intelligence, correction technologies represented by affective computing have gradually been initiated and promoted. While improving the quality and efficiency of correction and facilitating correction institutions to better perform their functions, they have also triggered problems and challenges such as algorithm black boxes and data security. Therefore, this paper examines the topic from the perspectives of technological regulation and human-oriented optimization, with the aim of achieving a qualitative change in correction work from “tool innovation” to “governance model leap”.