This paper presents a multi-subject collaborative data security governance system architecture that caters to various aspects of data security governance. It also introduces an attribute-based encryption technology that utilizes multi-subject data, specifically a multi-subject access control policy fusion algorithm scheme from a technical perspective. The neural network is chosen to build the prediction model. The particle speed is changed using the linear decreasing weight strategy. The improved particle swarm algorithm is used to set the initial weight coefficients and bias variables of the BP neural network. This creates the shared information security risk prediction model. In order to derive the maximum number of iterations of the data security risk prediction model, analyze the encryption and decryption overhead of the multi-owner access control strategy. Additionally, analyze the fluctuation amplitude of MAE and MRE that meets the error threshold range, selecting the virus attack industry for analysis. The probability of occurrence of information data risk points is assessed by analyzing the frequency of coding structure for data technology risk, data management risk, and data legal risk, which is combined with the interview data. The statistics show that the data management risk category, with a standard deviation of 32.64, is more volatile than the two core categories of data technology risk and data legal risk. The lack of stability exposes the data security of e-commerce platforms to risky turbulence.