Brussels has done it again. Barely a few months after approving one of the most ambitious regulatory frameworks in the world on artificial intelligence (and the first globally), the European Union has agreed to “simplify” its own law.
The so-called AI Act omnibus package introduces clearer rules, more flexible deadlines, and less burden for companies. It also adds new prohibitions, such as “nudification” applications. All in the name of competitiveness and security.
The Commission proposes, the Parliament toughens, member states negotiate … and, months later, the same system acknowledges that what was approved did not work as expected. This is not an isolated anomaly, but the usual method.
The case of artificial intelligence fits into a well-known pattern in Brussels. First comes legislation driven by regulatory ambition—often with more political will than technical understanding. Then come companies, national regulators and economic reality. And that is when phase two begins: correcting, refining, ‘simplifying.’
It happened with Schengen. The original design was based on an almost doctrinal premise: free movement without internal border controls except for limited exceptions. The 2015 migration crisis and terrorist attacks forced member states to reinstate controls for years. The legislation had to adapt to what was already happening. The rule did not anticipate politics; politics ended up rewriting the rule. And today, Schengen could collapse under the weight of the mass immigration that Brussels has promoted.
It also happened with the Growth and Stability Pact. The 3% deficit and 60% debt limits were the anchor of the euro. After the financial crisis, that framework was reformed through the ‘Six-Pack’and the ‘Two-Pack.’ A decade later, it was reformed again. Fiscal discipline shifted from rigid to flexible, and from flexible to negotiable on a case by case basis. The same system that aimed to impose automatic rules ended up acknowledging it needed “discretion.”
The digital sphere offers an even clearer example. The 1995 Data Protection Directive was replaced by the GDPR in 2016. And since then, the regulatory ecosystem has continued to expand: DSA, DMA, ePrivacy. Each new layer corrects, redefines, or displaces the previous one. The result is a framework where legal coherence competes with regulatory proliferation. And, of course, where data protection increasingly becomes the least relevant objective.
The carbon market (ETS) followed a similar logic. It was launched in 2005 with generous allocations that collapsed CO2 prices. Brussels responded with successive reforms, tightening the system, creating stability reserves, and expanding its scope. Each phase corrected the errors of the previous one. Each adjustment responded to a prior design failure.
The Common Agricultural Policy (CAP) is perhaps the most paradigmatic case. From chronic surpluses, it shifted to a system of decoupled payments and environmental conditionality. But that change was not the result of an initially sound vision, but of decades of corrections (and constant protests from farmers).
The same applies to financial services. MiFID gave way to MiFID II after the 2008 crisis exposed shortcomings in transparency and supervision. Banking rules have been rewritten several times under the CRD and CRR frameworks. Each crisis exposes the flaws of the previous system.
And now artificial intelligence. The AI Act was presented as the global standard—celebration by Thierry Breton included. But even before it is fully implemented, Brussels has had to adjust its architecture. More time, more clarity, less burden for SMEs. In other words, less than what was initially approved—and celebrated.
In practical terms, companies and citizens first operate under complex rules that are later softened. The transition cost is rarely recovered. Billions lost due to avoidable mistakes.
Why all this? Legislating generates institutional visibility. Correcting allows institutions to showcase responsiveness. Between the two, the European machinery justifies itself.
