Find and delete illegally exposed personal information using AI

Find and delete illegally exposed personal information using AI


Refer Report

The government is developing a system that can detect posts using artificial intelligence (AI) technology and retrieve personal information included in videos, such as faces.

The Personal Information Protection Committee announced on the 24th that it reported the ‘Measure to Strengthen Response to Exposure and Illegal Distribution of Personal Information’ containing these contents at the general meeting held on the 23rd.

This measure was designed to minimize exposure and illegal distribution of personal information on the Internet and prevent secondary damage.

The Personal Information Protection Commission decided to apply AI methods such as machine learning to the existing keyword-based detection method, reflecting the situation in which the exposure and illegal distribution of personal information is changing from text-centered to images and videos. To this end, we plan to develop and apply a conversation-type search term that can detect personal information exposed on over 4.7 million domestic and international websites.

It is also noticeable that the detection target is expanded from text to photos and images. In order to respond to the spread of deepfakes, it was decided to gradually upgrade the system that detects personal information included in videos, such as faces. The search scope of the ‘Find My Stolen Information’ service, which allows you to check whether personal information has been illegally distributed on the dark web, etc., will be expanded from current account information to email and phone numbers.

In order to quickly block posts exposing personal information, we will also actively utilize the ‘hotline cooperation system’ consisting of 15 major domestic and foreign portal sites and social media companies. Through this, we plan to shorten the deletion period of illegally distributed posts from 24.8 days last year to 18.9 days by next year.

Reporter Seungwoo Lee leeswoo@hankyung.com

Source: korean