EU's new AI law imposes penalties on ChatGPT and Claude! Copyright infringement and unauthorized training can incur fines of up to 7% of revenue.

The European Commission announced the "AI Practice Guidelines" on July 11, (General-Purpose AI Code of Practice), as a transitional mechanism to assist businesses in complying with the "AI Act." The content focuses on copyright protection and increasing the transparency of large AI models. It is expected to apply to large generative AI models such as ChatGPT and Claude starting in August, with violators facing fines of up to 7% of their revenue.

Prohibit the use of pirated data to train AI, protecting the rights of creators.

The European Commission stated that this code of practice requires developers not to use pirated content to train AI. If creators clearly indicate that they "do not wish their works or data to be used to train AI," developers must respect and exclude them.

In addition, once the AI-generated content is suspected of infringement, the operator must have a mechanism in place to address and rectify such issues.

Developers must provide complete AI documentation to implement transparency principles.

The Executive Committee added that AI model developers must provide the latest model usage documentation to regulatory bodies and third-party integrators, specifically disclosing how the AI operates and its limitations, and must implement the "principle of transparency."

This regulation mainly targets "General Purpose AI Models" (General Purpose AI), such as OpenAI's ChatGPT, Anthropic's Claude, etc., and is expected to be implemented starting in August.

The guidelines are voluntary in nature, and violators may face a penalty of up to 7% of annual revenue.

This guideline is of a "voluntary nature"; businesses can choose whether to sign it, with the aim of allowing companies to make adjustments in advance for the upcoming "AI Act." The Executive Committee emphasizes that businesses signing the guideline can obtain more legal protection and reduce the risk of future legal violations.

Violators of the "AI Act" can face fines of up to 7% of their annual revenue. For businesses developing advanced AI models, the maximum fine is 3% of their annual revenue. The guidelines were originally scheduled to be released in May this year, but the executive committee ultimately postponed it to July, and there are currently no plans to delay the implementation timeline of the act.

The overall guidelines are drafted by the European Commission, inviting representatives from AI laboratories, technology companies, academia, and digital human rights organizations to form a working group to jointly develop the content of the draft.

European and American tech companies have all expressed opposition.

However, this guideline immediately sparked backlash from tech companies like Meta and Alphabet, which criticized the initial draft for exceeding the original scope of the AI Act and placing additional burdens on businesses. In response, local European companies also voiced their protests.

Major semiconductor company ASML (ASML), Airbus, and AI startup Mistral AI, among others, called on the EU in early July to postpone the implementation of the "AI Act" for two years, advocating for a more innovation-friendly regulatory approach.

The issue of balancing creator rights and technological innovation still exists.

The legendary director Hayao Miyazaki of Studio Ghibli once stated:

"I really don't want to integrate AI into work; it's an insult to life itself. I feel that humanity is approaching the end of an era while simultaneously losing confidence in itself."

When AI models heavily rely on web-scraped data, and even use creators' works for training without authorization, it not only risks copyright infringement but also makes many artists feel that the value of their creations is diluted and their emotions are replaced. This time, the European Union has taken the lead in introducing AI regulations and practical guidelines. Although there are still controversies, it is undoubtedly an attempt to clarify the boundaries of "data licensing," "technology responsibility," and "respect for creation." In the future, achieving a balance between protecting creators, encouraging innovation, and preventing abuse will become a long-term challenge faced by legal systems and industries in various countries.

This article discusses the EU's new AI law targeting ChatGPT and Claude, imposing fines for piracy training and infringement up to 7% of revenue. It first appeared in Chain News ABMedia.