South Korea's AI Law: Why Startups Warn of Service Shutdowns Next Month

The AI framework act is scheduled to take effect on Jan. 22, 2026, calling for the establishment of a national AI committee, the formulation of a basic three-year AI plan and the imposition of safety and transparency requirements, including disclosure obligations for some AI systems, according to the sources.

If implemented as planned, South Korea would become the world's first country to enforce a comprehensive AI regulatory framework, reports Yonhap news agency.

The European Union was the first to pass AI-related legislation, but it plans to apply most of its rules starting August, with some provisions expected to be delayed until 2027 amid mounting pressure from businesses and intensifying global competition.

"Companies may not have sufficient time to prepare for the new rules, as the enforcement decree is expected to be finalised only shortly before the law takes effect due to procedural requirements," an official from the Korea Internet Corporations Association said. "This will be particularly overwhelming for startups."

A recent survey by Startup Alliance showed that 98 percent of 101 local AI startups said they have not established a response system to comply with the new law.

Of the respondents, 48.5 percent said they were unfamiliar with the law and unprepared, while another 48.5 percent said they were aware of it but ill-prepared.

"If the current implementation timeline is maintained, some companies may be forced to abruptly change or suspend services after Jan. 22," another industry official said.

"If regulations are too strict, companies will have stronger incentives to launch services overseas rather than at home," he added.

Industry watchers noted that such regulatory pressure is believed to be one reason a growing number of South Korean AI startups are considering Japan, which has adopted a softer, voluntary governance approach.

Particularly worrisome are the mandatory watermarking rules that would require AI-generated content to be labeled despite the need to curb deepfakes and other forms of misuse, they said.

"Even AI-generated contents often involve hundreds of people working to improve quality, but consumers may turn away once they are labeled as 'AI-generated,'" an official at an AI content company said.

"There are also ambiguities in the labelling requirements, and I don't think the views of those who fully understand the content creation industry and other relevant experts have been fully reflected," he added.

- IANS