The EU AI Act’s amendment process has two engines. The Council moves. The Parliament moves. They move separately, and eventually they meet. This week, Parliament moved.
IAPP confirmed that Members of the European Parliament reached a preliminary political agreement on EU AI Act amendments on or around March 11, 2026. Prior hub coverage addressed the Council’s amendment package, including the three-track structure and the nudification ban addition. This is a distinct institutional milestone: the Parliament side has now aligned on its position. Both institutions must agree for amendments to take effect.
Where the Amendment Process Stands
A preliminary political agreement at the committee level is not the end of the road. It’s a significant step, it means the relevant MEPs have reached internal alignment, but it still requires a full committee vote before proceeding to trilogue negotiations with the Council.
That committee vote is reportedly scheduled for March 18, 2026. Two days from the date of this writing. If the vote confirms the preliminary agreement, it advances the Parliament’s formal negotiating position. If it produces changes, the timeline shifts. Either way, March 18 is a concrete checkpoint.
The relationship to the Council’s position matters. Prior deep-dive coverage on this hub detailed the Council’s three-track amendment approach. Whether the Parliament’s position aligns with or diverges from the Council on the specific deadline extensions and generative AI provisions will determine how difficult, or straightforward, the trilogue process becomes.
The Reported Deadline Structure
The specific provisions require careful framing. This package’s verified source, the IAPP article, was partially accessible, meaning the headline is confirmed but the article body was not fully captured for independent verification. What follows reflects IAPP’s reporting and should be treated as reported, not final, until the March 18 vote produces official documentation.
According to IAPP, the MEP agreement reportedly extends two compliance deadlines:
Annex III high-risk AI systems: deadline reportedly extended to December 2, 2027. Annex III covers AI systems in high-stakes domains, biometrics, critical infrastructure, education, employment, law enforcement, migration, and administration of justice, among others. The practical scope is broad.
Annex I high-risk AI systems: deadline reportedly extended to August 2, 2028. Annex I references AI components embedded in products already governed by EU product safety legislation, medical devices, machinery, civil aviation, automotive. The longer runway here reflects the complexity of integrating AI Act requirements into existing product compliance regimes.
Both dates carry the same caveat: reported via IAPP, not independently confirmed from official EU text. The compliance-professional audience for this hub deserves that precision. Acting on unconfirmed deadline data in a regulated environment carries real risk.
The Generative AI Problem
The deadline extensions for high-risk systems point in one direction. The generative AI provision points in another.
According to IAPP reporting, the MEP agreement reportedly shortens the grace period for generative AI transparency requirements to three months. If confirmed, this is the provision that should concern operators of generative AI systems most immediately, and it runs counter to the direction of everything else in the amendment package.
A shortened grace period means less time to bring generative AI transparency obligations into compliance from the moment the amendments take effect. The practical implications depend on where an operator currently stands: companies that have been building toward compliance have different exposure than those that have treated generative AI transparency as a longer-term horizon item.
The provision’s exact scope, which generative AI transparency requirements specifically, and from which effective date the three-month clock runs, could not be confirmed from available source material. This is a known gap. Operators should flag this for immediate verification against the March 18 vote documentation.
Parliament vs. Council: What We Know and What We Don’t
Prior Council-side coverage established the three-track structure for the Council’s amendment approach. This deep-dive cannot make a direct Parliament-Council comparison on specific provisions because the Council’s positions on the exact deadline dates and the generative AI grace period weren’t mapped against the Parliament’s reported positions in this package.
What can be said: both institutions are moving in an amendment direction, both have reached internal alignment at different stages, and the trilogue process will determine where the two positions meet. Compliance programs should be scenario-planning for a range of outcomes, including the possibility that trilogue negotiations produce dates different from those currently reported.
What to Do This Week
The March 18 committee vote is the immediate action item. Same-day coverage will appear on this hub’s EU AI Act page and regulation feed.
For compliance program leads: the reported deadline structure, if confirmed on March 18, provides the planning horizon for Annex III and Annex I workstreams through late 2027 and mid-2028 respectively. That’s meaningful runway for high-risk AI system compliance, assuming the dates hold through trilogue.
For generative AI operators specifically: the reported three-month grace period compression is the provision that warrants immediate attention from legal counsel. If that provision survives the committee vote and trilogue, your transparency compliance timeline may be considerably shorter than current roadmaps assume.
Don’t anchor your compliance calendar on reported dates from a partially verified source. Watch March 18. Then update.
This piece builds on prior TJS coverage of the Council’s amendment package. For the Council-side context, see EU AI Act Amendment: What the Council’s Three-Track Package Means for Your Compliance Program.