EU Unveils Third Draft of AI Act Detailed Rules
Shift from Strict Regulations to More Lenient Terms
South Korea’s Ministry of Science and ICT Also Leans Towards Minimizing AI Basic Law Regulations
“A Bill That Encompasses Both Industry Development and Regulation is Needed”
The European Union (EU) has released the third draft of the detailed rules for the world’s first artificial intelligence (AI) regulation law, the “AI Act.” The draft includes more lenient language regarding AI copyright issues, signaling a shift from the previously strict regulatory stance. Amid intensifying global AI competition, there is growing criticism that South Korea’s AI Basic Law, passed at the end of last year, needs follow-up measures focused more on fostering rather than regulating AI.
According to foreign reports on the 13th, the EU Commission unveiled the third draft of the practical guidelines on the 11th (local time) to finalize provisions targeting providers of General Purpose AI (GPAI), which includes generative AI. The practical guidelines are specific instructions for enforcing the AI law. Since last year, the EU has been refining the draft rules by gathering feedback from stakeholders and companies. The provisions targeting GPAI providers are expected to take effect in August of this year.
These provisions impose obligations such as providing technical documentation, user manuals, compliance with copyright guidelines, and disclosure of summaries of data used for training these models. The AI Act categorizes services using large language models (LLM), like OpenAI’s “ChatGPT” and Meta’s “Llama,” as GPAI, making global generative AI companies subject to regulation. Violation of the AI Act could result in fines of up to 3% of annual revenue.
However, with the release of the new draft, there are indications that the previously “regulation-heavy” content has been relaxed. The EU stated that the latest amendment is “simplified in structure with more sophisticated commitments and actions” based on feedback from the second draft. Particularly noteworthy is the section on “AI training copyright,” which is a sensitive area for AI companies. Vague terms like “best efforts,” “reasonable measures,” and “appropriate actions” are used to ease copyright infringement in data collection for model training. The lack of precise regulation suggests uncertainty about the intention to apply the law.
Additionally, a previous draft provision requiring a single contact point and complaint handling mechanism for direct and quick copyright issue submissions has been removed. It has been modified to require designating a contact point for communication with affected rights holders and providing easily accessible information. The draft also specifies that if a copyright complaint is “clearly baseless or excessive, especially if repetitive,” action may be refused. This can be interpreted as allowing the dismissal of repeated complaints on similar issues by the same copyright holders.
The introduction of this draft suggests the EU, enacting one of the world’s toughest AI regulation laws, is stepping back before implementation. The EU’s revision of plans reflects concerns about falling behind in the global AI dominance race. Recently, the competition to develop AI models has expanded to a national rivalry level. The Trump administration announced the “Stargate Project,” investing approximately 730 trillion won in AI data centers and strengthening policy support. China has proposed a new paradigm by implementing high-performance AI models based on efficient algorithms, such as those represented by DeepSeek. Amid these developments, concerns exist that the EU’s excessive regulation may cause it to lag in global competition.
In response, there have been calls for South Korea’s AI Basic Law to follow up with measures focusing on development rather than regulation. South Korea passed the AI Basic Law, the second AI legal framework globally, following the EU, on December 26 last year. The AI Basic Law defines AI technologies that significantly impact users’ lives, safety, and fundamental rights as “high-impact AI” and stipulates obligations for the relevant AI operators. However, during the National Assembly passage, there was controversy over excessive regulatory authority. Especially, the industry has expressed dissatisfaction with a clause allowing the invocation of investigative powers based merely on civilian complaints or reports.
Regarding this, the Ministry of Science and ICT has decided to minimize regulations, considering the early stage of the AI industry, in drafting subordinate statutes for the AI Basic Law, scheduled for implementation in January next year. Science and ICT Minister Lee Jong-ho said at a monthly briefing on the 11th that “only the minimum necessary regulatory items will be included, considering the nascent stage of the AI industry” and that they would approach the interpretation of target and level of regulation cautiously. This aims to alleviate industry concerns over excessive regulation. The ministry plans to finalize specific details by gathering stakeholder opinions in the future.
Prof. Choi Byung-ho from Korea University’s AI Research Institute commented, “If Europe’s AI Act is too densely packed with regulations, South Korea’s AI Basic Law currently lacks specifics and remains abstract,” adding, “it’s crucial that the AI Basic Law aims to achieve both nurturing the AI industry and appropriate regulation.”