Transparency in the Labyrinths of the EU AI Act: Smart or Disbalanced?

Background: Complete transparency in artificial intelligence is impossible to achieve.[1] In the interdependent technological context, the scope of artificial intelligence transparency and the logic behind the values that outweigh transparency are unclear. Legislation on artificial intelligence, such as the European Union Artificial Intelligence Act (hereinafter the EU AI Act), tries to define the true meaning and role of AI transparency. Methods: The author applies doctrinal research and comparative analysis methods to assess AI transparency in the EU AI Act; a framework of distinct transparency zones is established. Doctrinal research helps to define the scope of transparency obligations and examine their limitations and interaction within the EU AI Act, while comparative analysis highlights inconsistencies, such as an unexplained difference between transparency duties in distinct zones or different requirements for open source and proprietary AI. Results and conclusions: The findings reveal a fragmented and uneven framework of artificial intelligence transparency in the EU AI Act, shaped by many exemptions, exceptions, derogations, restrictions, and other limitations. The zero-transparency zone (established by Article 2) is too broad, with much discretion given to stakeholders. In contrast, the basic transparency zone (set by Article 50) is too narrow, posing risks to fundamental human rights. The next zone, the moderate transparency zone (Chapter V), struggles with responsibility sharing between AI providers and downstream deployers. Meanwhile, the high transparency zone (provided in Chapter III) privileges law enforcement. Lastly, the hybrid transparency zone highlights complications in managing interactions between different risk-level AI systems.The author concludes that the EU AI Act is progressive but needs more fine-tuning to function as a coherent and solid transparency framework. The scales between public interest in artificial intelligence transparency, individual and societal rights, and legitimate interests risk being calibrated post-factum.