EU AI Regulation: 'Landmark agreement' on world's first comprehensive AI legislation

Marathon Negotiations Lead to Consensus

In an intense bout of negotiations, European Union policymakers have finally reached a preliminary agreement. The focus was on the key provisions of the Artificial Intelligence (AI) Regulation, which will govern the use and development of generative AI and foundation models. This latest round of discussions, involving the European Parliament, the European Council, and the European Commission - collectively known as the 'trialogue' - began on 6 December and ran continuously for almost three days.

The first part of these negotiations culminated in a consensus on the regulation of foundation models and issues of key importance to the creative industries. The second part of the discussions ended late on Friday 8 December, after another round of tough negotiations on enforcement and other key aspects of the law. However, it remains to be seen whether the regulation fully reflects the aspirations and legitimate concerns of the creative industries.

Key Points for the Creative Sector

Based on the information currently available, it appears that the new regulation will enforce transparency for all models. Companies developing AI models will have to provide a sufficiently detailed summary of the content used for the "training" process. This mandate is to some extent in line with the expectations of the creative sector. Another notable point is that AI-generated content will have to be labelled as such and be instantly recognisable. The law excludes open source and free models with publicly available parameters, but these models will still need to comply with copyright law and provide a detailed summary of the content used.

The Role of Industry Practices

Technology industry codes of good practice will serve as a complement to binding obligations until harmonized technical standards are established. If this proves to be a lengthy process, the European Commission has been given the power to intervene through delegated acts.

Does the Accord Meet the Expectations of the Creative Industries?

It remains challenging to provide a comprehensive and definitive answer to this question without studying the final version of the Act, which is on the verge of formal adoption by the EU co-legislators in 2024. Preliminary analysis suggests that while the Act includes several stipulations concerning the foundational models, it does not completely align with the maximalist agenda of the sector.

The Averted Risk

It is important to note that a significant risk was successfully averted. The regulation came dangerously close to excluding any obligations of transparency and accountability for companies developing foundational models, due to some last-minute concerns raised by France, later joined by Germany and Italy, which succumbed to the demands of their technology companies and called for self-regulation of the technology industry. The situation was finally resolved thanks to the strong position of the European Parliament, the compromise texts of the European Commission and the negotiating skills of the Spanish Presidency of the Union.

A Step Forward, Yet Vigilance is Crucial

Given the risk averted, the agreed regulation of generative AI, despite its shortcomings, can be seen as a positive step in the right direction. The sector will continue to closely monitor regulatory and practical developments, including delegated acts, enforcement, and standards development.

The Intersection of AI and IP rights

In analyzing the proposed regulatory framework, it is important to clarify that AI and IP rights are not mutually exclusive, but rather intertwined, each influencing the other in a sophisticated dance of technological progress and legal evolution. The purpose of regulation is to create the conditions and rules under which they can coexist, while ensuring a balance between technological development and the protection of fundamental rights, including copyright, in accordance with Article 17(2) of the Charter of Fundamental Rights of the European Union. The proposed regulation also underscores the importance of transparency, accountability, adherence to copyright law in AI's generative processes, including but not limited to the correct application of the relevant copyright exceptions.

In this context, it is important to note that, as a general rule, any use of works or other subject matter, including for the training of AI, requires an authorisation of the rightholders. The application of the relevant copyright exceptions, including for the purposes of so-called text and data mining, should always be assessed in accordance with the so-called three-step test, which outlines the general requirements for the permissibility of free use in principle. In the Bulgarian Copyright and Related Rights Act (CRRA), these conditions are laid down in Article 23: "The free use of works shall be permitted only in the cases provided for by law, provided that it does not interfere with the normal use of the work and does not prejudice the legitimate interests of the author". If the analysis finds that the three cumulative conditions for the permissibility of free use under the three-step test are met, then the examination of the applicability of specific exceptions provided for in the legislation, including those for text and data mining, as introduced in Articles 3 and 4 of Directive 790/2019 and already implemented in the Bulgarian CRRA, in force since 01 December 2023, could proceed. However, even under these exceptions, rightholders may choose to reserve their rights over their works or other subject matter to prevent text and data mining, except for the purposes of scientific research. Where the rights to opt out have been expressly reserved in an appropriate manner, providers of general-purpose AI models need to obtain an authorisation from rightholders if they want to carry out text and data mining over such works.

The need for regulation during the training phase is essential to ensure accountability and transparency, the proper application of exceptions, and the proper functioning of technology companies. Accountability is already a common practice among developers of generative AI systems. They have developed procedures for accessing and using training data.

Given the rapid and widespread use of copyrighted content in the generative processes of AI, which currently lacks adequate oversight, legislation becomes essential. Even companies that claim to prioritise transparency, fairness and respect for rights have begun to limit, eliminate or reduce their previous levels of transparency.

In conclusion, the intersection of AI and IP rights is a dynamic and complex area. The key to a sustainable future lies in maintaining a delicate balance that respects the rights of all stakeholders while fostering innovation and progress. Regulatory frameworks are therefore not only a legal necessity, but also an integral part of the harmonious and beneficial development of AI technologies.

The Creative Sector's Proactive Stance in Bulgaria

In May 2023, the creative sector, led by the Bulgarian Association of Music Producers/BAMP, joined the global Human Artistry campaign, which outlines the principles for the responsible use of AI to support human creativity. BAMP initiated a Common Position of the Bulgarian creative sector on the AI Regulation, supported by a total of 12 creative organizations. On the 5th of December 2023, the topic was brought to the public for the first time through the discussion panel "ARTIFICIAL INTELLECT. CREATIVITY. COPYRIGHT", jointly organized by BAMP and the Culture and Media Committee of the National Assembly, with attorney Desislava Mateva among the panelists.