EU rewrites parts of AI Act as Brussels bows to competitiveness pressure

The European Union is beginning to retreat from its original approach to artificial intelligence regulation. After years of presenting the AI Act as the global benchmark for trustworthy AI governance, Brussels is now under mounting pressure to reduce regulatory burdens and avoid slowing down Europe’s already fragile position in the global technology race.

Following overnight trilogue negotiations concluded on 7 May, negotiators from the European Parliament and the Council reached a provisional agreement on the so-called “AI Omnibus” package. 

The compromise introduces delays, exemptions and simplifications to key parts of the AI Act while formally maintaining the Union’s broader commitment to safety, transparency and fundamental rights protections.

Although EU institutions describe the package as a technical adjustment, the political significance is far greater. This is the first major rollback of EU digital legislation since the AI Act was adopted in 2024, and it reflects a broader shift in Brussels toward economic pragmatism and industrial competitiveness.

Delaying core obligations

The central change concerns high-risk AI systems. Under the agreement, obligations for standalone high-risk systems listed under Annex III will no longer apply from August 2026 as originally planned. Instead, they will enter into force on 2 December 2027. AI systems integrated into regulated products such as industrial machinery, medical devices and consumer equipment will only become subject to the rules from 2 August 2028.

The delays were strongly supported by several member states and industrial sectors which argued that companies faced growing uncertainty over technical standards, conformity assessments and overlapping legal obligations. 

European businesses have increasingly warned that the compliance framework surrounding the AI Act risked becoming excessively complex and expensive compared with regulatory environments in the United States and parts of Asia.

Germany secures industrial concessions

Germany played a particularly important role during negotiations. 

Berlin pushed heavily for revisions concerning industrial machinery, arguing that manufacturers faced duplicate obligations under both the AI Act and existing product safety legislation.

The final compromise reclassifies machinery legislation from Annex I-A to Annex I-B, reducing the risk of overlapping conformity procedures for companies operating in the sector.

The European Commission has also been granted additional powers to resolve future overlaps between the AI Act and sector-specific legislation through delegated acts expected by August 2027. The issue may appear technical, but it highlights a broader challenge for Brussels. As EU digital regulation expands across multiple sectors simultaneously, businesses increasingly complain about fragmentation, administrative duplication and legal uncertainty.

Mounting pressure

The political backdrop explains much of the sudden shift.

Over the past year, European policymakers have come under growing pressure from industry groups, national governments and economic analysts warning that Europe risks falling further behind the United States and China in artificial intelligence development. 

Mario Draghi’s widely discussed report on European competitiveness intensified those concerns by arguing that Europe’s regulatory culture was undermining innovation, investment and productivity growth.

American technology companies and senior US officials have repeatedly criticised European digital regulation as overly restrictive and hostile to technological development. 

Within Brussels itself, there is now growing recognition that Europe cannot simultaneously claim leadership in AI innovation while imposing compliance obligations that many companies consider unmanageable.

The Omnibus compromise therefore represents more than a procedural revision. It signals the beginning of a political recalibration inside the European Union.

That recalibration is particularly visible in the treatment of smaller companies. Simplified compliance rules initially designed only for SMEs will now also apply to so-called “small mid-caps”, medium-sized firms considered strategically important for Europe’s innovation ecosystem.

EU negotiators acknowledged that these firms often face regulatory costs similar to smaller businesses despite operating at larger scale.

Simplification not deregulation

Brussels is nevertheless attempting to preserve the political narrative that simplification does not mean deregulation without limits.

The agreement introduces explicit bans on AI systems used to create non-consensual intimate imagery and AI-generated child sexual abuse material. The measure responds to the rapid expansion of “nudification” tools capable of generating realistic fake sexual content using generative AI technologies.

Transparency obligations for AI-generated content also remain largely intact. Providers will still be required to register certain systems in the EU database even where exemptions apply. Watermarking and synthetic-content disclosure requirements have also been preserved, although implementation deadlines have been shortened from six months to three.

Some safeguards concerning sensitive personal data were strengthened during negotiations. 

The agreement restores the principle of “strict necessity” for the use of sensitive data intended to detect and mitigate algorithmic bias, an issue that had become increasingly controversial during discussions between Parliament and member states.

Simplification without full deregulation

The compromise now awaits formal approval by both the European Parliament and the Council before final adoption in the coming weeks.

For companies, the agreement offers additional time and greater legal clarity. For Brussels, however, it exposes a deeper structural problem inside the EU legislative process.

The AI Act only entered into force in August 2024, and large parts of the regulation are still not fully applicable. Yet less than two years later, the Union is already delaying and rewriting core sections of its flagship digital law. 

The speed of the revision process suggests that policymakers underestimated both the practical complexity of implementation and the economic impact of compliance obligations on European industry.

Europe reopens AI rulebook

The episode also illustrates a recurring feature of EU technology policy. 

Brussels frequently adopts ambitious regulatory frameworks before technical standards, enforcement mechanisms and industrial realities are fully aligned. The result is often years of renegotiation, implementation delays and political corrections once economic consequences become more visible.

Europe still insists that it wants to become the global leader in trustworthy AI. Increasingly, however, Brussels is also confronting another reality. In a global technology race moving at extraordinary speed, regulation alone cannot compensate for weaker investment, slower innovation cycles and fragmented industrial capacity.

The AI Omnibus package may temporarily reduce pressure on European businesses, but it also raises a more uncomfortable question for EU policymakers. 

If Europe’s flagship AI law already requires major simplification before its most important provisions even take effect, can the Union genuinely claim it has found a workable balance between regulating technology and remaining competitive in developing it?