Everything tech giants will hate about the EU’s new AI rules

The European Union is set to introduce a new framework aimed at enhancing transparency in the artificial intelligence sector. On Thursday, the EU unveiled a draft code of practice intended to guide major AI companies in adhering to the forthcoming AI Act, a groundbreaking legislation focused on copyright protections, transparency, and public safety. Although these guidelines will initially be voluntary for leading developers of 'general-purpose AI' starting August 2, enforcement of the AI Act will commence in August 2026. The European Commission has indicated that companies willing to comply with these guidelines could enjoy benefits such as reduced administrative burdens and greater legal certainty. In contrast, companies that choose to ignore these rules may find themselves facing more complex and costly compliance processes. The drafting of the AI Act included input from the AI industry, yet some firms have recently voiced concerns over the timing of the law's enforcement. They warn that stringent regulations could stifle innovation within the sector. One of the most contentious requirements involves a commitment from companies like Google, Meta, and OpenAI to refrain from using pirated content for AI training purposes. Historically, some AI firms have utilized unauthorized datasets, including pirated books, to train their models. For instance, Meta's defense of its practices drew criticism, as the company argued that individual books hold little value for training AI. However, the EU is insisting that tech companies establish dedicated teams and internal processes to handle complaints from rights holders, allowing them the option to exclude their works from AI training datasets. Furthermore, the new regulations will compel AI developers to disclose comprehensive information about their training data. This includes outlining the rationale behind critical design decisions and clarifying the origins of their datasets. Such transparency could illuminate the reliance of AI models on various data sources, including publicly available information, user data, third-party content, and innovative data types.