AI Legislation: Italy Sprints, Brussels adjusts the track and sends a message to Member States on national AI initiatives

Building on our previous insights, Italy's race to regulate the transparent and responsible use of AI through its Draft AI Bill has taken an unexpected turn after the European Commission's scrutiny.

While the Italian Government was busy discussing amendments to the Draft AI Bill in its sessions of 20 November and 27 November, the European Commission notified its observations on the Draft AI Bill, calling for alignment with the AI Act and urging the removal of any potentially conflicting or overlapping provision.

The new proposed amendments to the Draft AI Bill

As we previously noted, Italy is firmly committed to play an active role in shaping AI regulation.

Despite the recent publication of the AI Act, the Italian Government has presented (and is still working on) its own 26-article Draft AI Bill, aimed at regulating the transparent and responsible use of AI and laying down the principles on research, experimentation, development, adoption and application of AI systems and models.

In pursuit of this goal, new amendments were recently introduced to the Draft AI Bill. Among others, these updates propose a dedicated authority for AI and neuro-technologies, and strengthen the potential impact of the Draft AI Bill on workplace quality and AI training, and public health. Some amendments seem to focus on refining the scope of AI investment, narrowing it down to exclusively support AI (and, thus, excluding companies operating in the cybersecurity, quantum computing, telecommunications sectors) and establishing funds for the technological transfer of AI applications.

Interestingly, despite the extensive regulatory work being undertaken by EU institutions in the past few months, these amendments show Italy's willingness to continue to push its legislative efforts in the context of AI. In this respect, one standout amendment seeks to impose stricter requirements on providers of AI service used to generate video, audio, and textual content with the aim of safeguarding copyright. This proposal would mandate the identification and tracking of sources used by AI systems in their learning process, as well as identifiable markers for AI-generated content.¹ However, the AI Act only introduces transparency requirements in the context of manipulated content, while requiring the adoption of policies aimed at safeguarding copyright when developing a GPAI model.

The opinion of the European Commission on the Draft AI Bill

Against this background, the European Commission's intervention did not come unexpected.

In response to the notification of the Draft AI Bill by the Italian Government pursuant to the TRIS Directive, on 5 November 2024 the European Commission issued   detailed opinion (C(2024) 7814), stressing the need for alignment with the newly-introduced AI Act.

The key concerns span from the lack of a consistent definitional and referential issues between the AI Act and the Draft AI Bill, to more substantive issues such as the proposed restrictions on AI systems other than high risk ones, or the potential establishment of a double regulatory framework.

Some commentators have welcomed the EU Commission’s position, stating what was already obvious to them: the Draft AI Bill duplicates, contradicts, and even surpasses the AI Act. Others, also, expressed criticism, claiming that the Italian legislator is more focused on appearances than on introducing genuine, pioneering measures.

On this note, some argue that the Draft AI Bill reflects the Government’s intention to legislate extensively on all aspects of AI development. While the aim to provide clear regulation is understandable, this approach risks overburdening innovation and ultimately limiting Italy’s global competitiveness.

Harmonization is key

Overall, the European Commission's opinion underscores the importance of aligning Italy's Draft AI Bill with the AI Act. This alignment is indeed crucial to prevent regulatory fragmentation and ensure consistent implementation across Member states. Not surprising at all, considering the number of substantial issues raised by the European Commission on Italy's proposal.

Definitional and Referential Issues

The European Commission firstly recommends that the Draft AI Bill explicitly references the AI Act to ensure consistency and clarity. From another standpoint, it also notes that the national definition of "AI models" diverges from the AI Act and suggests adopting EU definitions directly instead of replicating them at the national level.

Data and Transparency Concerns

The opinion raises concerns about the term "critical data," recommending that its scope be limited to matters of national security.

In the context of healthcare AI, the Commission instead advises restricting information obligations strictly to the use of AI, cautioning against extending patient information requirements beyond those outlined in the AI Act – as through the Draft AI Bill such information obligations would instead extend to the "benefits, in diagnostic and therapeutic terms, from the use of new technologies" and "Information on the decision logic adopted".

Usage Restrictions

The European Commission calls for the removal of restrictions on AI systems other than high risk ones proposed by the Draft AI Bill in the context of learned professions, emphasizing the need for proportionality. This is not at all surprising, as the whole AI Act was adopted on the basis of a risk-based approach by carefully balancing all the interests at stake.

Authorities

The opinion deals with AI and authorities from a double perspective. On one hand, it once again calls for more consistency with the AI Act, allowing the use of AI systems within the judiciary provided that fundamental rights are not harmed. On the other hand, the European Commission stresses the need for independence of national AI authorities.

Content Identification and Manipulation

The controversial requirements for marking AI-generated content and "protecting" against such content are also targeted by the opinion. While the Draft AI Bill proposes using a prominent "AI" acronym to flag artificially generated content, the European Commission sees this as an overlapping and overregulation of existing obligations. Indeed, the AI Act already includes comprehensive provisions for (AI-generated) content transparency and user safety.

Sanctioning Framework

Seeing as the Draft AI Bill appears to be attempting to reinvent an entire regulatory wheel that already exists at the EU level – starting off with new definitions and ending with a delegation in the realm of sanctions – the European Commission also focuses on the sanctioning framework. In this context, the opinion reminds Italy that Article 99 of the AI Act already provides specific sanctions for infringements. It thus suggests harmonizing delegated powers in national legislation with the existing EU regulatory framework to avoid duplication or conflicts.

Strict guardrails for Member States' AI initiatives

All the amendments have been subject to a non-objection opinion by the members of the Italian Government, which stressed the need to take into account the opinion of the European Commission, despite not halting the adoption process of the Draft AI Bill.

The race for Italian AI legislation thus does not end there. But where are we headed?

In practice, it remains crucial to understand how the process will unfold, given that virtually all the key points of the Draft AI Bill have been criticized by the European Commission – and that the easiest way to avoid "overregulating" or "overlapping" with existing legislation would be for Italy to simply withdraw its proposal.

Zooming out, what really matters is the clear signal sent out by the European Commission's opinion: avoid overregulation and steering clear of regulating areas where the European institutions have invested significant effort to establish a  harmonized framework.

This reflects a broader commitment to fostering innovation while ensuring consistency of regulations remain consistent across Member states, supporting both the competitiveness of the EU and the integrity of its single market. A fragmented or overly restrictive approach risks undermining these objectives, creating unnecessary barriers for stakeholders across the AI ecosystem.

Ultimately, the European Commission’s stance demonstrates its determination to maintain firm control over the AI regulatory framework. This opinion serves as a warning to other Member states, also emphasizing the importance of the pre-emption principle of EU law – often forgotten by national governments. National legislators must thus avoid introducing autonomous rules or notions that diverge from the provisions of the AI Act, whose maximum harmonization nature must be respected, ensuring that national efforts align with the overarching goal of uniformity across the Union.

In this context, the Report on EU competitiveness by Mario Draghi underlines the urgent need for a balanced approach to regulation in the technology sector². The report, echoed by industry stakeholders, advocates for a harmonized regulatory framework to avoid fragmentation and ensure clear and proportionate rules.

Such an approach aims to align AI development with EU values while fostering competitiveness and positioning Europe as a leader in technological innovation and sustainable growth.

Authored by Giulia Mariuz, Alberto Bellan, Ambra Pacitti, Giacomo Bertelli, Cecilia Canova, and Anna Albanese.