Sixteen Capitals Missed the AI Deadline. On Friday They Meet to Discuss Europe's AI Ambitions

When the EU's 27 heads of state and government meet in Dublin on Friday for their first dedicated discussion on artificial intelligence, they will do so in capitals that have largely failed to implement the law they spent four years negotiating.

Sixteen member states have missed a deadline to appoint a national authority to enforce the AI Act, according to the Commission. Six countries, Austria, Belgium, Bulgaria, Croatia, Slovakia and Hungary, have not selected an authority at all. Ten others have made only a provisional decision that still requires formal adoption. The original deadline was August 2025.

The delays are not a technicality. National authorities are what turn obligations on paper into market surveillance, investigations and penalties. Without them, companies face a rulebook with no referee, and citizens have no route to complain about a system they believe has harmed them.

The letter

Sergey Lagodinsky, a German Green MEP, has written to the Commission asking it to open a formal procedure against member states for failing to meet their legal requirements under the AI Act.

“Stop blaming the EU for missing the race if you miss your own deadlines,” he said. “Europe's AI ambitions cannot be delivered if national capitals cannot get their act together.”

That argument will be awkward in the room on Friday, because it inverts the usual framing. For years, national governments have treated Brussels as the obstacle to AI competitiveness. The Commission's answer has been that the regulation exists and provides the certainty investors need. Both claims now run into the same fact: the enforcement layer is missing in most of the single market.

The investment that slipped twice

The more damaging delay is on the money. The EU has pushed back plans to support seven AI compute hubs across Europe, part of a €30 billion investment project announced at the end of October. The department running it has extended the deadline for consortia to express interest, citing the need to give member states time to complete national budgetary processes and submit financial commitments.

It is the second delay in a row. The initial November deadline had already moved to February.

“Apparently, some paperwork wasn't submitted on time,” posted Jan Kavalírek, who had a brief stint as the Czech government's AI envoy and is involved in the Czech bid. “I think it's very important that we keep some timeline and conditions that were presented at the beginning. Of course, this also applies to the member states, which should be able to keep up with the deadlines and deliver all necessary documents in time to Brussels.”

The pattern is now familiar: the EU legislates faster than it builds. Compute capacity, like enforcement capacity, depends on national administrations delivering on schedule, and the schedule has moved twice in three months.

What Dublin is supposed to decide

European Council President António Costa's invitation letter framed Friday's meeting around two questions: how to make Europe competitive in AI, and how to make it less dependent on the United States and China for the technology stack that underpins it.

That dependency has become a live political issue in Brussels. The budget impasse over the 2028-2034 framework has already blocked proposals linked to AI chips, the Arctic and agriculture, with policymakers unable to advance funding decisions while the size of the envelope is contested. Separately, an MEP has argued that a Starlink service cut-off led to deaths, part of a broader push to treat connectivity and compute as security infrastructure rather than markets.

Safety has climbed the agenda for reasons beyond regulation. A series of incidents involving rogue AI agents over the summer, more recent cyberattacks, and warnings from inside the industry about possible existential risk have made the case that guardrails are a precondition for deployment rather than a brake on it.

The asymmetry that will not go away

There is a defensible version of Europe's approach: regulate first, build trust, then invest in capacity that meets a known standard. There is also a version that has been failing for three years, in which the standard is set in Brussels and the capacity is delayed in 27 capitals.

Friday's meeting will produce language about urgency and sovereignty. The measurable test is narrower. How many of the sixteen capitals have named their authority by the time leaders sit down, and whether the seven compute hubs get a date that holds.

Both are entirely within member states' control, which is what makes the answer embarrassing either way.