In the early hours of 7 May 2026, after another marathon negotiation, the European Parliament and the Cypriot Presidency of the Council reached a provisional agreement on the Digital AI Omnibus, a text that amends the AI Act barely two years after its adoption and just weeks before most of its rules come into force. The text was then adopted by Parliament on 16 June, by the Council on 29 June, published in the Official Journal on 24 July and entered into force on 27 July 2026. It illustrates an increasingly visible tension: how to reconcile pioneering regulation with a technological lag behind the United States and China?
The AI Act: a regulatory ambition that clashes with reality
Finally adopted on 21 May 2024 and entering into force on 1 August 2024, the AI Act is the world's first comprehensive legal framework on artificial intelligence. It classifies AI systems according to their risk level and provides for fines of up to 35 million euros or 7 % of global annual turnover. Its application is gradual: prohibited practices, such as social scoring or subliminal manipulation, have been banned since February 2025; rules on general-purpose AI models have applied since August 2025; the most burdensome obligations, those for high-risk systems (biometrics, critical infrastructure, education, employment, law enforcement), were to apply on 2 August 2026.
It was this deadline that posed a problem. The harmonised technical standards, essential for demonstrating compliance, were not ready. In November 2025, the Commission therefore proposed the omnibus, as part of a simplification agenda fuelled by the Draghi report and the Budapest declaration of November 2024, which called for a 'simplification revolution'.
What the omnibus changes
- Postponement of high-risk obligations: they will apply on 2 December 2027 for autonomous systems listed in Annex III (biometrics, education, employment, law enforcement…), i.e. sixteen months later than planned, and on 2 August 2028 for systems integrated into already regulated products.
- Relief extended to small mid-cap companies: certain facilities reserved for SMEs now benefit companies with fewer than 750 employees and up to 150 million euros in turnover or 129 million in total balance sheet.
- Sensitive data to correct biases: their processing is possible to detect and correct biases, including outside high-risk, but only when strictly necessary, a requirement reinstated by the co-legislators.
- Industrial AI: machines are largely removed from the direct scope of the AI Act in favour of the Machinery Regulation, to avoid double regulation.
- A new prohibition: 'nudification' applications and the generation of child sexual abuse material, as of 2 December 2026.
- Other adjustments: national regulatory sandboxes postponed to 2 August 2027 and marking of AI-generated content required by 2 December 2026 for systems already on the market.
The new prohibition echoes the investigation opened on 12 January 2026 by the British regulator Ofcom against X, following the dissemination of sexualised images generated with its AI Grok, including of children.
The European dilemma: regulate first, innovate later?
The omnibus is part of a broader questioning of the European approach. The Union was the first to regulate AI comprehensively, but its companies struggle to compete with American and Chinese giants, which benefit from larger markets and far greater investments. The paradox recalls that of the GDPR: Europe sets global rules, without having the champions who would benefit from them.
To correct this imbalance, the Commission launched the 'AI Continent' plan in 2025. According to EuroHPC, 19 'AI Factories', centres combining supercomputers, data and expertise, are being deployed, complemented by 13 antennas. The InvestAI initiative targets 200 billion euros, and calls have been launched for AI 'gigafactories', according to Euronews.
A sufficient simplification?
Reactions to the agreement show the difficulty of the exercise: no one is truly satisfied. CCIA Europe, which notably represents large American digital companies, welcomed the postponement while calling it 'the bare minimum', and denounced 'minimal' improvements. Conversely, BEUC, the European consumer organisation, believes through its Director General Agustín Reyna that the omnibus 'rolls back key consumer protections adopted barely a year ago'.
Kai Zenner, advisor to MEP Axel Voss, sums up the problem for the IAPP: the agreement gives companies more time and largely removes machines from the scope of the law, but these 'small steps' do not solve structural problems; the European digital acquis needs, in his view, a major overhaul, particularly in terms of legal coherence and enforcement. Companies must indeed navigate between the AI Act, the GDPR, the Digital Services Act, the Digital Markets Act, the Data Act and the Machinery Regulation.
Europe faces a strategic choice
The stakes go beyond compliance deadlines. The Union has bet that by setting trust standards, it would attract players seeking a predictable framework, as with the GDPR. But the United States favours deregulation and massive private investment, while China combines state support and targeted regulation. Between the two, Europe is trying to chart a third path: a 'trustworthy' AI, rooted in fundamental rights, but which must also be competitive.
The omnibus is a first adjustment, which implicitly acknowledges that the initial ambition had to be tempered. As Isabelle Roccia, Europe Director of the IAPP, notes:
« I can't help but wonder what the AI Act would have looked like if, two years ago, legislators had given themselves a little more time to produce a finished text. » (Isabelle Roccia, IAPP, translated from English)
Two years after its adoption, Europe is already amending its own rules. Sign of regulatory maturity or admission that, in the AI race, legislating quickly does not mean legislating well? The answer will depend less on the text than on its implementation: technical standards finally available, national authorities with adequate resources, and the ability to actually enforce the prohibitions already in force.
Sources: Council of the EU, provisional agreement on the AI omnibus, 7 May 2026; Council of the EU, final adoption, 29 June 2026; European Parliament, vote of 16 June 2026; Regulation (EU) 2026/1744, OJ of 24 July 2026; European Commission, 'AI Omnibus enters into force', 27 July 2026; European Commission, fact sheet on small mid-cap companies; Ofcom, investigation into X and Grok, 12 January 2026; IAPP, 'EU agrees to amend AI Act, clarifies overlap with machinery rules', 7 May 2026; Isabelle Roccia, IAPP, 'A view from Brussels', 7 May 2026; CCIA Europe, 7 May 2026; EuroHPC JU, AI Factories and antennas, February 2026; Euronews, InvestAI and gigafactories, 8 September 2026.
