A Single Investigation Could Cripple AI Startups
Clear Responsibility Needed for AI Watermarking
Government to Finalize Enforcement Decree by End of June
The government has announced that it may ease the so-called "fact-finding investigation" clause?widely considered one of the most controversial provisions of the Artificial Intelligence (AI) Basic Act set to take effect in January next year?through subordinate enforcement decrees. The fact-finding investigation, stipulated in Article 40 of the Basic Act, has raised significant concerns within the industry. This is because the government can only initiate an investigation upon receiving a report if AI-generated content lacks an AI watermark, if risks are not properly disclosed, or if the source of data used for AI training is not revealed. In response, the government has stated that it will include language in the draft enforcement decree allowing for exceptions to fact-finding investigations. However, industry voices argue that clear and specific criteria for these exceptions are necessary, suggesting that there is still a long way to go before a final consensus can be reached.
A Single Investigation Could Cripple AI Startups
According to government and industry sources on April 22, the Ministry of Science and ICT recently held a meeting to gather feedback from SMEs and startups regarding the draft enforcement decree of the AI Basic Act. During the meeting, the ministry reportedly clarified that "exceptions to fact-finding investigations may be allowed in the enforcement decree."
Participants emphasized that establishing concrete criteria for exceptions is more important than simply allowing for exceptions themselves. One industry representative at the meeting stated, "The government must specify the criteria for exceptions in the enforcement decree to reduce uncertainty, but it has not done so. This leaves room for public officials to interpret 'exceptions' arbitrarily, so the controversy over the problematic clause remains unresolved."
Another startup representative commented, "If we become the target of a malicious complaint, we will be forced to divert manpower and time to respond to the investigation. Unlike large corporations, startups do not have dedicated legal teams, so this process alone could destroy a startup in an instant."
Clear Responsibility Needed for AI Watermarking
Another major point of contention at the meeting was the lack of clarity over who is responsible for marking AI-generated images, text, or videos as AI creations. Article 31 of the AI Basic Act, which covers the "obligation to ensure transparency in artificial intelligence," merely states that businesses must indicate AI-generated content. However, the enforcement decree should specify "who ultimately bears responsibility for such labeling," and participants pointed out that this is currently missing.
A startup representative at the meeting cited an AI video editing startup as an example, saying, "Under the current draft enforcement decree, it could be interpreted that the startup is obligated to inform the film production company of AI usage, but the production company bears no responsibility to notify the audience." He added, "Even if the startup applies an AI watermark, if the production company removes it, the startup would still be held fully liable." The startup industry collectively argues that, to avoid such unreasonable situations, the enforcement decree should assign labeling responsibility to the service provider closest to the end user.
The Ministry of Science and ICT plans to finalize the enforcement decree of the AI Basic Act by the end of June. Detailed guidelines and official notices will be released alongside it. To prevent misunderstandings that could arise from the wording of the enforcement decree alone, the ministry is also considering including separate examples or model cases in the guidelines.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
