Personal Information Protection Commission Supports Strengthening Privacy Protection for AI Training Data

[Asia Economy Reporter Eunmo Koo] The Personal Information Protection Commission announced on the 23rd that it is supporting the Ministry of Science and ICT’s AI training data construction project, carried out through the Korea Internet & Security Agency (KISA), to ensure compliance with personal information-related regulations. Since 2017, the Ministry of Science and ICT has been building large-scale AI training data and making it available through the ‘AI Hub.’

To promote the successful implementation of the Digital New Deal project by creating a safe data utilization environment, the Personal Information Protection Commission signed a business agreement with the Ministry of Science and ICT in November last year. During the AI training data construction phase, KISA reviewed 63 sample types for the presence of personal information and the possibility of re-identification. As a result, additional de-identification measures were requested for some image and video data where individuals and vehicle numbers were identifiable, as well as for some text data.

The Ministry of Science and ICT and the National Information Society Agency (NIA), together with specialized de-identification companies, conducted a final inspection of 65 types of AI training data comprising over 180 million records that might contain personal information and proceeded with additional pseudonymization. These pseudonymized AI training data are scheduled to be sequentially released by the end of this month. The AI Hub will operate an online window for three months, allowing users to identify and provide feedback on any insufficiently pseudonymized data, which will then be promptly improved.