When the EU Artificial Intelligence Act was finalized in late 2023, European Commission President Ursula von der Leyen hailed it as a law that would “transpose European values to a new era.” Three years later, her language had shifted considerably. The same legislation now, in her words, needed to create “a simple, innovation-friendly environment.”

That’s a significant rhetorical journey in a very short time. And the amendments finalized in May suggest the reality matches the tone.

What Changed — and What Didn’t

The headline changes: the implementation deadline for high-risk AI systems has been pushed from August 2026 to December 2027. Requirements for AI integrated into products already covered by sector-specific regulations — cars, medical devices, machinery — have been delayed until August 2028. And critically, AI used in industrial settings, including car manufacturing and semiconductor production, has been carved out entirely from high-risk requirements.

That last carve-out matters. It was a priority for German Chancellor Friedrich Merz, whose coalition governs Europe’s largest industrial economy and whose legacy manufacturers — automotive, chemicals, machinery — were not enthusiastic about navigating a separate AI compliance regime on top of existing product safety rules. Germany got what it wanted.

The revisions also banned “nudifier” apps and AI-generated child sexual abuse material — narrower, genuinely uncontroversial changes that passed without industry pushback.

What didn’t change: the rules for general-purpose AI models from companies like OpenAI, Anthropic, and Mistral. Those are still coming in August. That distinction is worth sitting with.

The Innovation Argument — Does It Hold?

Supporters of the amendments argue the EU was moving too fast, that regulation was strangling European competitiveness, and that the changes bring the bloc into step with the global AI race. This framing has been embraced enthusiastically by major American technology lobbying groups, including AmCham EU and the ACT Association.

Critics counter that this argument is backwards. The AI Act had barely been enforced — no fines had been issued, most key provisions hadn’t yet taken effect. Europe’s AI competitiveness problem, they argue, is better explained by a lack of venture capital and a fragmented single market than by regulatory burden. The law hadn’t really been tested yet.

There’s a more pointed version of this critique too: that the simplification push was largely the result of sustained lobbying by American technology firms who saw the high-risk AI framework as a potential competitive disadvantage. The joint letter from American and European industry groups in April 2026, calling for “a clear, simple and innovation-friendly implementation of the AI Act,” does rather invite the question of whose interests were being served.

The Brussels Effect, Under Pressure

What’s interesting about this moment is the strategic contradiction at its heart.

The EU spent years arguing that its regulatory approach would become a global standard — the so-called Brussels Effect, where EU rules get adopted elsewhere because it’s easier for multinationals to comply with one standard than many. GDPR is the canonical example.

The AI Act rewrite complicates that story. Europe is simultaneously exempting its own industrial sector from the toughest rules while keeping strict requirements on general-purpose AI models — the category where American companies dominate. One reading: Europe is building space for its own champions while holding foreign competitors to a higher standard. Another reading: Brussels is learning that the Brussels Effect only works when you’re not in a geopolitical fight with the companies you’re trying to regulate.

Whether the AI Act remains a global template or becomes a cautionary tale about regulatory capture and geopolitical naivety may be the defining question for European tech policy over the next five years. The August GPAI deadline will be the first real test.

Source: The Parliament Magazine