Strasbourg trilogue: deal in sight, with one open question

The second and likely final political trilogue on the AI Omnibus took place in Strasbourg on April 28. Parliament rapporteurs Arba Kokalari (EPP) and Michael McNamara (Renew) faced Council negotiators under the Cypriot Presidency, and reports from the session indicate the institutions converged on most outstanding issues. A provisional political agreement is widely expected.

What’s now agreed in principle:

• Fixed high-risk postponement dates: December 2, 2027 for stand-alone Annex III systems and August 2, 2028 for AI embedded in regulated products under Annex I
• A targeted ban on “nudifier” AI systems
• Strict-necessity guardrails for processing sensitive personal data for bias correction

The principal divergence: Parliament wants to move all Annex I-A product categories into Annex I-B, integrating AI Act requirements horizontally into sectoral legislation (medical devices, toys, radio equipment, machinery). The Council didn’t include a comparable provision, and civil society lobbied strongly against the change.

For where institutions stood going in, see the A&O Shearman briefing from April 24 and the Ropes & Gray alert on what to expect as negotiations conclude.

“The AI Omnibus is entering its final stretch. Whatever the outcome, the real work, for businesses, and legal teams, starts the moment the ink dries.”

Giulia Stancampiano, Director of Legal (Privacy & Tech), iubenda

Annex I-A: why this technical question matters

The Annex I-A debate is the defining fault line of the AI Omnibus. At stake: whether the AI Act keeps direct scope over AI built into regulated products like medical devices and toys, or whether those products are governed only by their existing sectoral rules.

CDT Europe’s April AI Bulletin calls this the sharpest open point. CDT and 32 other civil society organizations signed an open letter calling on co-legislators to preserve Section A and maintain the AI Act’s direct scope. On the other side, industry actors have lobbied for extending Article 50 transparency deadlines and reintroducing a registration exemption for self-assessed non-high-risk systems.

The Bulletin also flags a coalition of US-based organizations urging EU institutions to resist external pressure to water down AI Act requirements. How this gets resolved will shape who gets enforced against, and under which framework, for some of the most consequential AI use cases in Europe.

Also this week

A LEOsphere tracking note flagged uneven Member State progress on designating market surveillance authorities under the AI Act, with France and Spain ahead and several others still in consultation. The Commission’s AI Act Service Desk and Single Information Platform are now live. Meanwhile, the Kaizenner AI Act Omnibus tracker remains the deepest public archive of documents from all three institutions. And a Jacques Delors Centre analysis noted that the non-AI Digital Omnibus (GDPR, ePrivacy, cybersecurity, data acquis) still has no formal trilogue start date, with disagreement within Parliament on file responsibilities.

Worth watching

None of this is law yet. The Omnibus is still a proposal in trilogue, and anything could change before formal adoption. That said, a few things are worth keeping on your radar:

• Confirmation of the political agreement. A deal is widely expected, but watch for the formal announcement and any last-minute changes.
• The Annex I-A outcome. This will determine whether the AI Act keeps direct authority over AI in regulated products, or hands that role to sectoral rules.
• The data/non-AI Omnibus track. GDPR, ePrivacy, and data-related provisions are still upstream of trilogue. The timing and shape of those negotiations is the next big unknown.