Brussels postpones stricter AI rules for 16 months amid political and business pressure

The European Commission announced on Wednesday a 16-month delay in the implementation of the stricter rules of its Artificial Intelligence Act, moving its entry into force from August 2026 to December 2027. The change, included in the “Digital Omnibus” regulatory packageThe measure arrives in Brussels amid strong pressure from major technology companies, European SMEs, and US President Donald Trump. The Commission argues that it is a necessary simplification, while civil society organizations warn of an anticipated regulatory dismantling.

A high-impact postponement amidst regulatory tension

The European Commission surprised everyone on Wednesday by announcing a significant delay in the implementation of its Artificial Intelligence ActThe most ambitious AI regulations in the world. The package “Digital Omnibus” postpones until December 2027 the obligations applicable to the systems of high risk, a change that completely alters the planned regulatory timetable.

The Vice President for Technological Sovereignty, Henna VirkkunenHe defended the decision, arguing that European companies “They are hampered by excessive bureaucracyand that it was necessary to adjust the deadlines “without reducing safeguards.” The Commission argues that the delay is due to a technical bottleneck: The necessary harmonized standards do not yet exist. to properly apply the security and transparency obligations provided for in the law.

Despite this reasoning, the measure has generated concern among those who saw the AI ​​Act as the world's leading regulatory referenceapproved just a few months ago after arduous internal negotiations. For these sectors, the announcement represents "the first sign of withdrawal" in a field that Europe aspired to lead.

The change also comes in a complex geopolitical context. Community sources admit that some of the pressure came from major US technology companies and the administration of Donald Trump, who has repeatedly denounced European regulations as a threat to "Western innovation".

SMEs relieved, large technology companies satisfied

The 16-month delay affects systems used in biometric identification, granting of credit, staff pick, critical infrastructures, education., legal assistance o public securityThese are considered the areas with the greatest social risks. Companies that develop or deploy these systems will have more time to certify their algorithms, demonstrate security, document data, and ensure human oversight.

Brussels also proposes a relaxation of technical requirements for SMEsThis point has been widely welcomed by European business associations. According to the Commission, these simplifications could generate a minimum savings of 225 million euros per year, with an estimated maximum of 433 million if Member States fully implement the exemptions.

Another relevant change is the elimination of the direct obligation of suppliers to promote AI literacy among the end users; now that responsibility falls on the member states, who will have to develop national training programs.

The decision has been enthusiastically received by European People's Partywhich hailed it as “a crucial boost for industrial competitiveness.” It has also been well received by the Computer & Communications Industry Association (CCIA), which includes Apple, Google, Amazon and Meta, whose spokespeople indicated that it is about “a promising first step”, although they demanded “a deeper reform”.

The position of large European companies follows the same line: ASML, Airbus, Mistral AI, Mercedes-Benz and Siemens They had previously requested a two-year pause in the implementation of the AI ​​Act, arguing that the regulations could put the mainland industry at a disadvantage compared to the US and China.

Civil society denounces an unprecedented capitulation

While the industry is celebrating the delay, digital rights organizations have reacted with unusual forcefulness. 127 civil organizations, including the Austrian NGO noybThey described the proposal as “the biggest rollback of fundamental rights in the EU’s digital history”.

The Green MEP Kim van Sparrentak She was even more direct: “Reopening technology legislation is a dismantling of the system in favor of Trump.”He stated, accusing the Commission of giving in to US and business pressure.

To Agustín Reyna, Director General of the European Consumer Organisation (BEUC), the postponement represents a “weakening of privacy standards"which will primarily benefit "large non-European technology companies", the only ones capable of absorbing regulatory uncertainty and adapting their global systems to different regulatory frameworks.

Also, the analyst from the Center for Democracy & Technology, Laura Lázaro, warned that the reform “unacceptably weakens the once prestigious EU digital framework”, built over a decade to protect citizens from opaque and discriminatory algorithmic systems.

In the European Parliament, progressive parties fear that the delay will open the door to further changes that lower the law's requirements even more, especially in sensitive areas such as biometrics, emotional recognition, and the automation of decision-making.

An uncertain horizon for the AI ​​Law

The “Digital Omnibus” still needs to be approved by the Council of the EU and the European Parliamentwhere tense negotiations are expected. Some member states—led by France and Germany—have been lobbying for months to relax the law. Others, such as Spain and the Nordic countries, have defended the need to maintain high regulatory standards.

Technology governance experts point out that the EU risks losing international leadership just as countries like China or the United States are rapidly advancing in the regulation and commercialization of advanced AI models. “Europe is diluting its regulatory advantage before even implementing it.”"comments a researcher from the European Digital Policy Centre."

The delay also poses operational challenges: if the technical standards aren't ready now, it's not clear they'll be ready by 2027 either. Furthermore, the rapid evolution of generative models could render certain articles of the law obsolete before they even come into effect.