Naver goes all-out to detect deepfake search results

2024.02.22 10:22:02 | 2024.02.22 11:19:24

[Photo by MK DB]이미지 확대

[Photo by MK DB]

South Korean internet giant Naver Corp. is planning to attach a warning label to search results for deepfake keywords that may manipulate information ahead of the general elections on April 10.

According to information technology (IT) industry sources on Wednesday, Naver will unveil measures to block harmful deepfakes next week, which can distort information, defame people, and cause damage.

The measures are broadly categorized into two ?distribution and production, according to the sources.

On distribution, Naver is considering displaying a kind of guidance message when users search for specific keywords that may be linked to harmful deepfakes. Naver is currently carrying out internal discussions on which keywords to classify as harmful deepfake search values, the sources said.

When it comes to production, Naver has restricted deepfake creation within its CLOVA X, an artificial intelligence (AI) chatbot based on its own large-scale language model, HyperCLOVA X.

Currently, it provides a limited image editing function to some users, which is a form of multimodal, but blocks responses to questions with safety issues such as facial synthesis.

Naver‘s measures are in line with growing calls from the public that platform companies take an active role in blocking deepfakes.

The Korean ruling and opposition parties, in the meantime, passed a bill amending the Public Official Election Act to completely ban election campaigns using deepfakes.

With the amendment, which took effect on January 29, 2024, the use of deepfake videos, photos, and audio for one’s election or the defeat of an opposing candidate is prohibited 90 days before the election date, with penalties including up to 7 years of imprisonment or fines ranging from 10 million won ($7,500) to 50 million won.

Nonetheless, deepfake posts are growing in number.

According to an investigation by the National Election Commission from January 29 to February 16, there were 129 posts violating the Public Official Election Act through election campaigns using deepfakes.

By Ko Min-suh and Yoon Yeon-hae

[ⓒ Pulse by Maeil Business Newspaper &, All rights reserved]