[The Editors' Verdict] The Significance of the Declaration: "AI Cannot Use News Content for Training Without Permission"

At a recent event where the government unveiled a draft guide on copyright for the use of generative artificial intelligence (AI), it was noteworthy that the government stated a "separate contract is necessary" regarding the copyright exception of 'fair use.' This marks the first time the government has taken a stance amid the ongoing clashes between major tech companies like portals aggressively entering AI training and the media industry over whether news content qualifies as 'fair use.'

At the 'Seoul Copyright Forum' hosted by the Ministry of Culture, Sports and Tourism on the 26th, the Copyright Commission under the government stated, "In situations where the applicability of fair use is unclear, unauthorized use of copyrighted works is considered to have the potential for (copyright) infringement." They further urged, "Service providers must enter into separate contracts with rights holders before using copyrighted works to secure lawful usage rights." Since there are no court precedents on the fair use clause under copyright law, the government cannot make legal judgments, so this is intended as a precautionary measure to prevent legal risks through 'contracts.'

Fair use is an exception to copyright protection. Article 35, Paragraph 5 of the Copyright Act lists four major factors: 'purpose and character of use,' 'type and usage,' 'proportion and importance of the used part,' and 'impact on the market or value,' stating that "copyrighted works may be used if it does not conflict with normal usage methods and does not unjustly harm the legitimate interests of the author." The Copyright Act enumerates cases such as trials, political speeches, and school education that are not subject to copyright restrictions, and leaves other matters to be judged by courts under fair use. Recently, news content, which has become a point of sharp contention, is also included under fair use due to the absence of separate regulations.

It is also significant that the announcement mentioned not only the "establishment of contract mechanisms" but also that "unauthorized use of copyrighted works may constitute (copyright) infringement." This can be seen as the government cautiously revealing its position amid the ongoing clashes between big tech and creators over interpretations of current law, without the ability to make judicial decisions. A Ministry of Culture, Sports and Tourism official met at the forum said, "The atmosphere has definitely shifted toward protecting creators."

This is closely related to the government's changed perception of AI copyright. The general assessment is that the government's stance began to shift somewhat around last summer. Until then, there was a strong tendency to argue that copyright restrictions should be loosened to foster the AI industry, but after creators' voices began to emerge, the focus shifted toward 'copyright protection.'

Though it is just a single statement from the government, it is expected to have a considerable impact on the ongoing copyright discussions. It strengthens the creators' claim that their works cannot be subject to 'fair use' and increases the likelihood of following a 'separate contract' procedure. A cautious atmosphere is also sensed at the National Assembly level. Although a bill proposing to add 'reproduction and transmission for information analysis' to copyright exceptions, allowing users to utilize content with peace of mind, has been submitted, discussions are expected to be difficult for the time being.

The government has operated a copyright system improvement working group since early this year to produce the AI copyright guide. If this year is about setting the direction, next year will focus on concrete work targeting stakeholders such as rights holders' compensation methods and usage permissions. It is highly likely that whether to apply copyright protection will be considered on a content-by-content basis. There is also a need to deliberate on how to judge content with ambiguous copyright protection and how to assess data value. There is still a long way to go.