The decisive week — 8 days to the AI Act trilogue

The most consequential week for EU AI governance since the Act came into force begins today. The Commission, Council, and Parliament have broadly aligned on extending high-risk deadlines to 2 December 2027. What they have not aligned on — and what civil society now argues is indefensible — is Parliament's proposal to move Annex I product-embedded AI systems largely outside the Act's direct scope. This is not a scheduling debate. It is an argument about what the Act is for.

On April 15, ARTICLE 19 and forty partner organisations published a joint open letter to the European Parliament and Council urging rejection of amendments in the AI Omnibus that, they argue, weaken the AI Act beyond its original intent. Their core objections: the proposal removes transparency requirements for high-risk AI, leaves people in the EU without adequate protection from biometric identification and AI used in schools, and was drafted without standard impact assessment or public consultation procedures.

One day later, CDT Europe submitted formal feedback raising a distinct but related concern: Parliament's proposal to move Section A of Annex I under Section B would push high-risk AI systems covered by sectoral product safety legislation largely outside the AI Act's direct scope. This covers machinery, toys, radio equipment, medical devices, and more — the systems where AI most directly touches consumer safety.

The Gen AI Summit Europe 2026 concluded Saturday evening in Valencia — the second edition of what has become southern Europe's most significant generative AI gathering. Over two days at Innsomnia, the organisers targeted 500+ professionals from 200+ companies across 15+ industries for technical tracks on autonomous agents, inference architectures, frontier models, and scalable deployment.

What is unusual is what the published programme itself reveals. Alongside the technical tracks sat a substantial dedicated stream on AI Act compliance, data governance, OWASP Top 10 for LLMs, and risk assessment frameworks. This is the engineering-compliance convergence made visible: the builders in the room now need to know whether August 2026 still holds, because their product timelines depend on the answer.

On April 8, Arthur Mensch — CEO of Europe's most valuable AI company — urged European governments to redirect public procurement toward domestic technology providers. The argument: absent deliberate preference, Europe's €14 billion AI champion and every company like it will continue buying American infrastructure, and European sovereignty becomes a marketing slogan with no operational teeth.

Mensch's phrase was deliberate: "vassal state." Earlier in March he had proposed a 1–5% levy on foreign AI model providers serving the European market — a proposal most capitals have not endorsed. But as the Omnibus trilogue opens, the procurement argument has a different audience. Member states are drafting AI Factory partnerships. The European AI Continent Action Plan turns one year old. The gap between industrial policy and actual purchasing behaviour is starting to show.

When Cyprus took the rotating Council Presidency for H1 2026, the Digital Omnibus on AI was identified as a priority file. By March, the Cypriot team was coordinating technical-level alignment between Council working groups and Parliament rapporteurs. By April, they had set the political trilogue for 28 April in Strasbourg — a date the entire European AI ecosystem is now organising its calendar around.

Small Presidencies are sometimes described as passive — as administrators running the European machine between larger capitals. Cyprus is not being passive. It has built the Omnibus compromise architecture: fixed deadlines (2 December 2027 for Annex III, 2 August 2028 for Annex I), retained the Council's proposed ban on AI-generated CSAM, and left the most contested question — Annex I's scope — for the trilogue itself to resolve.

Cyprus — a country of fewer than 1.2 million people, the third-smallest EU member by population — is, for the next seventy-one days, the Council of the European Union's Presidency. Every Tuesday, Cypriot ministers chair Council formations. Every working group on the Digital Omnibus on AI is being steered through Cypriot chairs. When the trilogue opens on 28 April in Strasbourg, it is Cyprus — not France, not Germany — that will formally represent the Council across the negotiating table from Parliament and Commission.

The Cypriot approach has been characteristically small-state: technical, coordinative, patient. Where Parliament has proposed dramatic Annex I restructuring, Cyprus has pressed for fixed deadlines both sides could live with. Where civil society has demanded restoration of deleted transparency requirements, Cyprus has quietly kept some of the weaker Council positions on the table as leverage.

What to watch this week: the Cypriot Presidency's willingness to concede on Article 49(2) transparency in exchange for retaining the fixed timelines. It is the most obvious landing zone for a compromise — and therefore the most politically explosive one.