Back to Blog

Dublin down on AI

November 2026ByArq Editorial Team

Nine things EU leaders should agree at their summit

irish and eu flags
Blog PostCommentaryStrategic Indispensability

The time for talking has long passed as EU leaders meet in Dublin on Friday 13th November for an informal summit on artificial intelligence.

AI is the defining issue that will shape most of the other questions on the table, from growth and jobs to security. Europe knows what steps it must take to strengthen its position in the global AI ecosystem: more compute, more energy, secure access to the most capable models and the ability to test them independently.

Arq Foundation’s experts have listed nine initiatives that could be adopted in Dublin and initiated within weeks and months with sufficient political will. Rather than focusing on headline-grabbing initiatives, EU leaders must adopt a set of actions to build the plumbing for transformative AI in Europe.

1. Keep AI on the leaders’ agenda

Informal summits can set a direction: at Versailles in 2022, leaders agreed to phase out Europe’s reliance on Russian gas, oil and coal. Dublin is a chance to set one for AI.

President Costa has already said leaders will return to AI in 2027, including at their next informal retreat. They should go further and return to the issue at each formal European Council, in a similar manner to the European Defence agenda which has featured as a regular item on European Council agendas since the leaders’ February 2025 defence retreat.

This is especially acute as AI policy falls between telecoms and competitiveness ministers, impeding the joined-up approach needed to cut across issues like energy, compute, trade and export controls.

These issues cannot be left to languish in dozens of official-level EU working groups; this challenge demands full-bodied political forward momentum, progress-checking and direction at the cross-cutting level that only Presidents and Prime Ministers can provide.

2. Secure access to the most capable AI

European access to the most capable US frontier models is no longer a given. In June, US export controls cut users outside the United States off from some of the most capable American models for almost three weeks, and in September the White House asked US labs to hold new models back from the UK’s AI Security Institute until Washington had reviewed them.

Leaders should discuss how they can leverage European bargaining power to ensure the most capable models available commercially at the time are not switched off in Europe by another government’s decision.

Concretely, the European Council should ask the European Commission to urgently map Europe’s strengths in the AI supply chain - firms like ASML, Zeiss and Siemens Energy - and protect them.

As many middle powers such as the UK, Canada, South Korea and Japan are in a similar situation to Europe, measures should be coordinated at the highest political level both to seek guaranteed access, and to prepare a common response if access were to be cut by the US Administration again.

3. Build compute to compete, with a clear rulebook

The EU has around five percent of global AI computing capacity, compared to three quarters held by the United States.

Leaders can commit to urgently plan for faster permits, direct power grid connections and clear rules for community consultation, water and energy use and environmental protection.

The EU accounts for about 15 percent of the world economy, and its share of global compute should move towards that. Countries do not need to wait for the EU's proposed Cloud and AI Development Act to start narrowing the gap: the first compute from the zones it plans is not expected before 2029, and its fast track covers permits but not grid connections.

4. Stop AI companies marking their own homework

Companies building the most capable AI models decide themselves whether they are safe to release. This summer, AI agents being tested by OpenAI escaped their test environment and broke into systems at Hugging Face, which the UN’s independent scientific panel on AI called an early warning of one way humans could lose control of AI systems. In July the European Commission promised an EU capacity for independent testing by 2027, and in September President von der Leyen said she wants to work on it with Canada and the UK, which bring technical expertise in safety testing.

EU leaders should request a dedicated budget line, which the Commission’s plan lacks, and a firm start date, starting joint testing with partners before the most powerful models are placed on European markets; and a shared system to report serious incidents.

5. Give Europe’s cyber defenders the best AI first

In October, hackers turned up the temperature of fridges in hundreds of French supermarkets to as high as 50°C overnight, spoiling food and starting fires. On 2 November a second wave hit more than 100 stores in Belgium, Ireland, the Netherlands, Sweden and other countries. The attackers did not need AI: they got in through old controllers connected to the internet with default passwords. AI is making weaknesses like these cheaper and faster to find, for attackers and defenders alike.

In July the Commission promised a European blueprint, due by the end of this year, for giving organisations structured access to advanced AI for cybersecurity. Leaders should hold it to that date, give national cybersecurity agencies and operators of critical infrastructure priority access, and fund the planned campaign to secure the open-source software Europe depends on beyond its pilot.

6. Build tools to check any future AI deals

Calls for a global AI agreement are growing. President von der Leyen and the leaders of Member States including Germany, Ireland, the Netherlands and Spain have signed the Finnish-Norwegian call for control of frontier AI models, which asks governments to explore an international institution able to set standards and enable verification, and the UK has made an agreement a priority of its 2027 G20 Presidency. Europe can make itself indispensable by developing the technology that could check whether any deal is being kept, within the data centres themselves.

EU leaders should turn President von der Leyen’s pledge to work with the UK and Canada on verification into funding for the currently nascent technology that would check how data centres are being used, and invite the remaining Member States to sign the Finnish-Norwegian call. The first step towards any global deal is building the tools to verify and build trust.

7. Build the expertise to oversee and adopt AI

AI development far outpaces states’ ability to keep up. EU leaders should agree that each country will present a plan within 12 months on bringing into government the skills and staff needed to oversee AI and support its adoption in government.

In addition, the European Commission should be encouraged to alter the terms upon which it hires top talent in the AI Office, to ensure global talent is attracted on compensation packages that are more commensurate with frontier companies.

8. Back breakthroughs and be their first customer

European inventors must look abroad for finance or to scale their European inventions. Europe loses both the invention, its eventual tax revenues, and any say over its evolution.

The European Innovation Council, the EU's flagship programme for breakthrough innovation, has been held back by Commission hiring and procurement rules and too little independence.

In the next EU research programme - FP10 - starting in 2028, leaders should push for programme managers with their own budgets and room to take risks, similar to the UK’s ARIA or the US agency DARPA.

Leaders should also make sure the new Innovation Act lets governments be the first customer for European innovations, and lets national agencies such as Germany's SPRIND and the Netherlands' NADI keep buying research quickly.

Leaders should also invite EU countries to set up agencies with a clear mission and the freedom to move fast, to develop technologies, such as early warning systems for new disease outbreaks, that keep Europe's defences ahead of new threats from AI.

9. Prepare European science for AI

AI is changing the way science is done. Research funders like the European Research Council will struggle ever-more to deal with the flood of applications that AI facilitates. AI systems for science are trained mostly on published research, which over-represents successes and leaves out failed experiments that also have value.

EU leaders should task the European Commission with testing innovative new ways of funding science at a speed and scale that matches the AI transformation. They should also reward researchers who publish negative results as their data is still useful for AI training.