ISSN 2663-0583 (Online)
ISSN 2663-0575 (Print)
  • Home
  • Transparency in the Labyrinths of the EU AI Act: Smart or Disbalanced?
"

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

Summary: 1. Introduction. – 2. Research Methodology. – 3. Overview of AI Transparency in the EU AI Act. – 3.1. Particularities of AI Transparency in the EU AI Act. – 3.2. Limitations of AI Transparency. – 4.Transparency Zones. –4.1. Zero Transparency Zone. – 4.2. Basic Transparency Zone. – 4.3. Moderate Transparency Zone. – 4.4. High Transparency Zone. – 4.5. Hybrid Transparency Zone. – 5. Conclusion.

Full Text:

DOI

HTML Abstract About Author(s) References Reviews Українською

  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.

[1] Mona Sloane and others, ‘Introducing Contextual Transparency for Automated Decision Systems’ (2023) 5 Nature Machine Intelligence 188, doi:10.1038/s42256-023-00623-7.

Abstract

 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.

[1] Mona Sloane and others, ‘Introducing Contextual Transparency for Automated Decision Systems’ (2023) 5 Nature Machine Intelligence 188, doi:10.1038/s42256-023-00623-7.

About Authors

 Gintare Makauskaite-Samuole 

Institute of International and EU Law, Mykolas Romeris University, Lithuania

gintare.samuole@gmail.com

https://orcid.org/0009-0007-7777-1873

Corresponding author, responsible for writing, reviewing and editing this article.

Competing interests: No competing interests were disclosed.

Disclaimer: The author declares that her opinion and views expressed in this manuscript are free of any impact of any organisations.

References

 

1.    Andrews M, ‘Emerging Best Practices for Disclosing AI-Generated Content’ (Kontent.ai, 30 August 2023) <https://kontent.ai/blog/emerging-best-practices-for-disclosing-ai-generated-content/> accessed 18 December 2024.

2.    Broda E, ‘Agentic Mesh: The Future of Generative AI-Enabled Autonomous Agent Ecosystems’ (Medium, 6 November 2024) <https://medium.com/towards-data-science/agentic-mesh-the-future-of-generative-ai-enabled-autonomous-agent-ecosystems-d6a11381c979> accessed 18 June 2024. 

3.    Buick A, ‘Copyright and AI training data–transparency to the rescue?’ [2024] Journal of Intellectual Property Law and Practice jpae102, doi:10.1093/jiplp/jpae102.

4.    Colonna L, ‘The AI Act’s Research Exemption: A Mechanism for Regulatory Arbitrage?’ in Gill-Pedro E and Moberg A (eds), YSEC Yearbook of Socio-Economic Constitutions 2023: Law and the Governance of Artificial Intelligence (Springer Cham 2023) 51, doi:10.1007/16495_2023_59.

5.    Davies P, ‘Are AI Companies Complying with the EU AI Act? A New ‘LLM Checker’ Can Find Out’ (EuroNews, 16 October 2024) <https://www.euronews.com/next/2024/10/16/are-ai-companies-complying-with-the-eu-ai-act-a-new-llm-checker-can-find-out?utm_source=substack&utm_medium=email> accessed 18 December 2024.

6.    Draghi M, The Future of European Competetiveness: Part A: A Competiteveness Strategy for Europe(European Commission 2024). 

7.    Ebers M, ‘Truly Risk-Based Regulation of Artificial Intelligence: How to Implement the EU’s AI Act’ [2024] European Journal of Risk Regulation 1, doi:10.1017/err.2024.78.

8.    Fanni R, ‘Why the EU Must Now Tackle the Risks Posed by Military AI’ (CEPS, 8 June 2023) <https://www.ceps.eu/why-the-eu-must-now-tackle-the-risks-posed-by-military-ai/> accessed 18 December 2024.

9.    Fraser H and Bello y Villarino JM, ‘Acceptable Risks in Europe’s Proposed AI Act: Reasonableness and Other Principles for Deciding How Much Risk Management Is Enough’ (2024) 15(2) European Journal of Risk Regulation 431, doi:10.1017/err.2023.57.

10. Frattone C, ‘Reasonable AI and Other Creatures: What Role for AI Standards in Liability Litigation?’ (2022) 1(3) Journal of Law, Market & Innovation 15, doi:10.13135/2785-7867/7166.

11. Geng Y, ‘Transparency for What Purpose?: Designing Outcomes-Focused Transparency Tactics for Digital Platforms’ (2024) 16(1) Policy & Internet 83, doi:10.1002/poi3.362. 

12. Haataja M and Bryson JJ, ‘What Costs Should We Expect from the EU’s AI Act?’ (Center for Open Science, 27 August 2021) SocArXiv 8nzb4, doi:10.31235/osf.io/8nzb4.

13. Haresamudram K, Larsson S and Heintz F, ‘Three levels of AI transparency’ (2023) 56(2) Computer 93, doi:10.1109/MC.2022.3213181.

14. Hauer MP and others, ‘Quantitative Study About the Estimated Impact of the AI Act’ (arXivLabs, 29 March 2023) arXiv:2304.06503 [cs.CY], doi:10.48550/arXiv.2304.06503.

15. Heintz M, ‘Watermarking for AI Text and Synthetic Proteins’ (Towards Data Science, 7 November 2024) <https://towardsdatascience.com/watermarking-for-ai-text-and-synthetic-proteins-fighting-misinformation-and-bioterrorism-fd45be625dfe> accessed 17 November 2024.

16. Holst L and others, ‘The Impact of the EU AI Act’s Transparency Requirements on AI Innovation’ (19th International Conference on Wirtschaftsinformatik (WI), Würzburg, Germany, September 2024) <https://eref.uni-bayreuth.de/id/eprint/90313/> accessed 17 November 2024.

17. Hosain MT and others, ‘Path to Gain Functional Transparency in Artificial Intelligence with Meaningful Explainability’ (2023) 3(2) Journal of Metaverse 166, doi:10.57019/jmv.1306685.

18. Jedličková A, ‘Ethical Considerations in Risk Management of Autonomous and Intelligent Systems’ (2024) 14(1-2) Ethics & Bioethics 80, doi:10.2478/ebce-2024-0007.

19. Kazim E and others, ‘Proposed EU AI Act–Presidency Compromise Text: Select Overview and Comment on the Changes to the Proposed Regulation’ (2023) 3 AI Ethics 381, doi:10.1007/s43681-022-00179-z.

20. Kelly J and others, ‘Navigating the EU AI Act: A Methodological Approach to Compliance for Safety-Critical Products’ (arXivLabs, 26 March 2024) arXiv:2403.16808 [cs.AI], doi:10.1109/CAI59869.2024.00179.

21. Khan T and Srikumar M, ‘Developing General Purpose AI Guidelines: What the EU Can Learn from PAI’s Model Deployment Guidance’ (Partnership on AI, 26 November 2024) <https://partnershiponai.org/developing-general-purpose-ai-guidelines-what-the-eu-can-learn-from-pais-model-deployment-guidance/> accessed 18 December 2024.

22. Kieseberg P and others, ‘Controllable AI - An Alternative to Trustworthiness in Complex AI Systems?’ in Holzinger A and others, (eds), Machine Learning and Knowledge Extraction: 7th IFIP TC 5, TC 12, WG 8.4, WG 8.9, WG 12.9 International Cross-Domain Conference, CD-MAKE 2023, Benevento, Italy, 29 August – 1 September 2023 (LNCS 14065, Springer Cham 2023) 1, doi:.1007/978-3-031-40837-3_1.

23. Kiseleva A, Kotzinos D and De Hert P, ‘Transparency of AI in healthcare as a multilayered system of accountabilities: between legal requirements and technical limitations’ (2022) 30(5) Frontiers in artificial intelligence 879603, doi:10.3389/frai.2022.879603.

24. Klyman K, ‘Acceptable Use Policies for Foundation Models’ (2024) 7(1) Proceedings of the AAAI / ACM Conference on AI, Ethics and Society 752, doi:10.1609/aies.v7i1.31677.

25. Lisinska J and Castro D, ‘The AI Act’s AI Watermarking Requirement Is a Misstep in the Quest for Transparency’ (Center for Data Innovation, 9 July 2024) <https://datainnovation.org/2024/07/the-ai-acts-ai-watermarking-requirement-is-a-misstep-in-the-quest-for-transparency/> accessed 19 July 2024.

26. Nannini L, ‘Habemus a Right to an Explanation: so What? – A Framework on Transparency-Explainability Functionality and Tensions in the EU AI Act’ (2024) 7(1) Proceedings of the AAAI / ACM Conference on AI, Ethics, and Society 1023, doi:10.1609/aies.v7i1.31700.

27. Novelli C and others, ‘AI Risk Assessment: A Scenario-Based, Proportional Methodology for the AI Act’(2023) 3 Digital Society 13, doi:10.1007/s44206-024-00095-1.

28. Perez P, ‘Treating AI Agents as Personas: Introducing the Agent Computer Interaction era’ (Medium, 5 November 2024) <https://medium.com/user-experience-design-1/treating-ai-agents-as-personas-6ef0135bdcad> accessed 18 December 2024.

29. Powell R, ‘The EU AI Act: National Security Implications’ (CETaS Explainers, 31 July 2024) <https://cetas.turing.ac.uk/publications/eu-ai-act-national-security-implications> accessed 18 December 2024.

30. Prifti K and others, ‘From Bilateral to Ecosystemic Transparency: Aligning GDPR’s Transparency Obligations with the European Digital Ecosystem of Trust’ in Kuhlmann S and others (eds), Transparency or Opacity: A Legal Analysis of the Organization of Information in the Digital World (Nomos 2023) 115, doi:10.5771/9783748936060-115. 

31. Siegmann C and Anderljung M, The Brussels Effect and Artificial Intelligence: How EU Regulation Will Impact the Global AI Market (Center for the Governance of AI 2022).

32. Sloane M and others, ‘Introducing Contextual Transparency for Automated Decision Systems’ (2023) 5 Nature Machine Intelligence 187, doi:10.1038/s42256-023-00623-7.
Varošanec I, ‘On the Path to the Future: Mapping the Notion of Transparency in the EU Regulatory Framework for AI’ (2022) 36(2) International Review of Law, Computers & Technology 95, doi:10.1080/13600869.2022.2060471.

Reviews for article

Add a Review

Reviews are moderated and will be published after verification!
All comments not related to reviews of the article will be deleted!

АНОТАЦІЯ УКРАЇНСЬКОЮ МОВОЮ

 

Дослідницька стаття

 

ПРОЗОРІСТЬ У ЛАБІРИНТАХ ЗАКОНУ ЄС ПРО ШІ: РОЗУМНО ЧИ НЕЗБАЛАНСОВАНО?

 

Гінтаре Макаускайте-Самуоле

 

АНОТАЦІЯ

Вступ: Повної прозорості у сфері штучного інтелекту досягти неможливо.[1] У взаємозалежному технологічному контексті обсяг прозорості штучного інтелекту та логіка цінностей, які переважають над прозорістю, залишаються незрозумілими. Законодавство про штучний інтелект, наприклад, Закон Європейського Союзу про штучний інтелект (далі - Закон ЄС про ШІ), намагається визначити справжнє значення і роль прозорості ШІ.

Методи: Автор застосовує доктринальне дослідження та методи порівняльного аналізу для оцінки прозорості штучного інтелекту в Законі ЄС про ШІ; встановлюються межі щодо чітких зон прозорості. Доктринальне дослідження допомагає визначити обсяг зобов’язань щодо прозорості та вивчити їхні обмеження та взаємодію в Законі ЄС про штучний інтелект, тоді як порівняльний аналіз висвітлює неузгодженості, такі як непояснена різниця між обов’язками щодо прозорості в окремих зонах або різні вимоги до відкритого та пропрієтарного ШІ.

Результати та висновки: Результати дослідження розкривають фрагментарну та нерівномірну структуру прозорості штучного інтелекту в Законі ЄС про ШІ, сформовану багатьма винятками, відступами, лімітами та іншими обмеженнями. Зона нульової прозорості (встановлена ​​статтею 2) є надто широкою і надає значну свободу дій зацікавленим сторонам. На противагу цьому, основна зона прозорості (встановлена ​​статтею 50) є надто вузькою, що створює ризики для фундаментальних прав людини. Наступна зона, зона помірної прозорості (Розділ V), має проблеми з розподілом відповідальності між постачальниками штучного інтелекту та тими, хто розгортає технології. Тим часом зона високої прозорості (описана в Розділі III) надає перевагу правоохоронним органам. Нарешті, гібридна зона прозорості підкреслює труднощі в управлінні взаємодією між системами ШІ з різними рівнями ризику.

Автор робить висновок, що Закон ЄС про штучний інтелект є прогресивним, але потребує доопрацювання, щоб функціонувати як цілісна та надійна система прозорості. Шкала між суспільним інтересом до прозорості штучного інтелекту, індивідуальними та суспільними правами та законними інтересами ризикує бути відкаліброваною постфактум.

Ключові слова: прозорість ШІ, Закон ЄС про штучний інтелект, зони прозорості, регуляторні обмеження, інновації проти прозорості.


[1] Mona Sloane and others, ‘Introducing Contextual Transparency for Automated Decision Systems’ (2023) 5 Nature Machine Intelligence 188, doi:10.1038/s42256-023-00623-7.

 

Keywords

  • AI transparency, EU AI Act, transparency zones, regulatory limitations, innovation vs. transparency

Publication history

  •  

    Date of submission - 22 Dec 2024

    Date of acceptance - 06 Feb 2025

    Date of Online First publication: 26 Mar 2025

    Whether the manuscript was fast tracked – No

    Number of reviewer report submitted in first round – Three

    Number of revision rounds – 1 round, revised version submitted 06 Feb 2025

    Technical tools were used in the editorial process:

    Plagiarism checks - Turnitin from iThenticate https://www.turnitin.com/products/ithenticate/

    Scholastica for Peer Review https://scholasticahq.com/law-reviews

    Managing editor – Mag. Bohdana Zahrebelna. Section Editor – Prof. Tetiana Tsuvina. English Editor – Julie Bold. Ukrainian language Editor – Lilia Hartman.

How to cite it?

  • Makauskaite-Samuole G, ‘Transparency in the Labyrinths of the EU AI Act: Smart or

    Disbalanced?’ (2025) 8(2) Access to Justice in Eastern Europe 1-31 ˂https://doi.org/

    10.33327/AJEE-18-8.2-a000105˃ Published Online 26 Mar 2025


     This research is part of the project The Jean Monnet Center of Excellence ‘European Fundamental Values in Digital Era’, 101085385 – EFVDE – ERASMUS-JMO-2022-HEI-TCH-RSCH. Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or EACEA. Neither the European Union nor the granting authority can be held responsible for them.