The European Union (EU) has released the third draft to finalize the detailed rules of the world's first artificial intelligence (AI) regulation law, the 'AI Act.' The draft shows a retreat from its previous stringent regulatory stance, featuring numerous softened expressions regarding AI copyright. As global AI competition intensifies, there are calls for South Korea's AI Basic Law, passed at the end of last year, to focus more on fostering rather than regulation.
According to foreign media on the 13th, the EU Commission announced the third draft of a practical guideline aimed at finalizing provisions for general purpose AI (GPAI) providers on the 11th (local time). Practical guidelines refer to specific instructions for enforcing the AI Act. Since last year, the EU has released two drafts of practical guidelines and is refining them based on feedback from stakeholders and corporations. The provisions targeting GPAI providers are expected to take effect in August this year.
This provision imposes obligations on GPAI providers, including providing technical documents and user manuals, compliance with copyright guidelines, and disclosing summaries of the data used for learning. The AI Act defines GPAI as AI services using large language models (LLMs) like OpenAI's 'ChatGPT' and Meta's 'LLaMA,' suggesting that global generative AI corporations will be subject to regulation. Violations of the AI Act could result in fines of up to 3% of annual revenue.
However, the release of this draft has led to criticisms that the previously 'regulation-heavy' content has been relaxed. The EU stated, 'Based on feedback on the second draft, the latest revision features a 'streamlined structure with sophisticated commitments and measures' compared to earlier drafts.' Particularly noteworthy is the section on 'AI learning copyrights,' which includes vague terms like 'best efforts,' 'reasonable measures,' and 'appropriate actions' to mitigate copyright infringement when collecting data for model training. The ambiguity of the regulations suggests an uncertain intent in their application.
Additionally, a provision requiring a single point of contact and complaint handling route to ensure 'direct and rapid communication' when copyright issues arise has been removed from the previous draft. It has now been changed to state that contacting and easily accessing information for those affected must be designated. It has also been specified that copyright complaints can be dismissed if they are 'clearly unfounded or excessive, especially if they are repetitive.' It could be interpreted that rights holders can ignore complaints if they repeatedly raise issues that have previously been filed.
The announcement of this draft is seen as a step back for the EU, which is set to enforce the world's strongest AI regulation laws. There is a sense of crisis that altering the previous plans may lead to falling behind in the global AI hegemony competition. The recent competition in AI model development is expanding into a national rivalry. The Trump administration announced the 'Stargate Project,' which involves an investment of about 730 trillion won in AI data centers, indicating a commitment to policy support. China has demonstrated a new paradigm by implementing high-performance AI models based on efficient algorithms, represented by DeepSeek. In this context, there is concern that excessive EU regulation on the AI industry could result in falling behind in global competition.
In this regard, there are calls for South Korea's 'AI Basic Law' to also focus on fostering rather than regulation. South Korea's AI Basic Law was passed on December 26 last year, making it the second in the world after the EU. The AI Basic Law defines AI technologies that significantly affect users' lives, safety, and fundamental rights as 'high-impact AI' and stipulates obligations for relevant AI operators. However, the AI Basic Law faced controversy over excessive regulatory authority during the parliamentary process. Particularly, the AI industry has raised objections to a provision that allows authorities to initiate investigations based solely on simple complaints or reports.
In relation to this, the Ministry of Science and ICT has set a direction to minimize regulations considering the initial stages of the AI industry in formulating subordinate regulations for the AI Basic Law, which is set to be implemented in January next year. Minister Yoosang-im stated during a monthly briefing held on the 11th that the direction for the AI Basic Law would include only the minimum regulatory requirements, noting, 'I will approach interpretations regarding regulatory targets and levels with caution.' This is in an effort to dispel industry concerns about excessive regulation. The Ministry of Science and ICT plans to finalize specific details by gathering opinions from stakeholders in the future.
Choi Byung-ho, a professor at Korea University's AI Research Center, stated, 'While the EU's AI Act includes numerous detailed regulations, the current content of South Korea's AI Basic Law lacks specificity and is highly abstract.' He emphasized that 'it is important for the AI Basic Law to strike a balance between fostering the AI industry and implementing appropriate regulations.'