Ireland pulled off something important at the International AI Summit. Representatives of 24 EU member states adopted the Dublin AI Declaration.

It deserves more attention.

Its full title is the Dublin Declaration on Trustworthy, Safe and Sovereign Artificial Intelligence. It sets seven priorities:

  1. Advance European AI safety and resilience.
  2. Build trustworthy applied AI capability.
  3. Strengthen European AI sovereignty.
  4. Invest in AI infrastructure and skills.
  5. Ensure the sustainability of AI.
  6. Lead through public-sector innovation.
  7. Deepen European and international cooperation.

This is a political declaration rather than a new law. That is part of its value. Europe already has the world’s most developed AI regulation. The declaration addresses the harder question of what Europe must now build.

Its first four commitments are particularly important. Europe needs the technical capacity to test advanced AI, detect incidents and defend critical systems. It needs to help businesses and public institutions apply AI safely. It needs control over its data, knowledge and essential technology. It also needs the compute, infrastructure and skills to make any of this real.

The remaining commitments make this a broader programme. AI infrastructure must be sustainable. Public procurement can create demand for trustworthy and interoperable European technology. International cooperation can turn shared principles into practical projects.

The beginnings of a wider alignment

The Dublin AI Declaration also sharpens an emerging international alignment among countries that do not control the largest frontier AI labs.

The clearest precursor came from Finland and Norway. On September 21, they published A Call for Control of Frontier AI Models, backed by leaders from Australia, Canada, Ireland and countries across Europe, Africa, Asia and the Middle East. The European Commission also endorsed it. The statement remains open and now has the backing of 29 countries, including 15 EU member states. It calls for mandatory testing before frontier models are deployed, independent evaluation, shared reporting of serious incidents, international verification and exploration of a new international institution to coordinate controls when dangerous capabilities emerge.

That statement did not come from all 27 EU member states, as has sometimes been suggested. But it was a substantial cross-regional call for international oversight, issued just 23 days before the Dublin AI Declaration.

Other partnerships are turning the same broad position into industrial policy. In April, Canada and Finland agreed to cooperate on sovereign AI capacity, access to compute, responsible infrastructure and reduced strategic dependence. In September, Canada and Germany backed LawZero, a concrete investment in safe-by-design AI research and sovereign computing infrastructure. Finland and Singapore then issued a joint statement on cooperation in the age of AI, supporting common standards, resilient technology ecosystems and broadly accessible benefits.

This is not a formal coalition. It is a growing network of bilateral partnerships and aligned initiatives. But a common position is becoming visible: countries should build enough domestic and allied capability to exercise meaningful control over AI; safety must become a technical capability rather than a promise; and the benefits should spread across society rather than accrue to a handful of companies.

Together, these countries could provide a counterweight to the US-China race for AI supremacy. They do not need to build the world’s biggest model. They can cooperate to make powerful AI safe, useful and subject to democratic control.

The overlap with the case made before the summit

Read side by side, the overlap with the case I made before the summit in Europe can regulate mad and bad AI. Now we must build the ability to make it safe is striking. The declaration’s first four priorities follow the same argument: Europe must move beyond regulating AI and build the safety, applied capability, sovereignty and infrastructure needed to use it on our own terms.

The overlap goes further. The op-ed proposed automated cyberdefence, realistic testing environments, open and interchangeable technology, European infrastructure, and public procurement based on safety and sovereignty. It also argued that Ireland could convene this work across Europe and the international AI industry.

That argument was never mine alone. My op-ed distilled a vigorous discussion across Ireland’s AI community, involving researchers, founders, officials, advisers and many others. The Dublin AI Declaration turns that discussion into a shared European agenda, and connects it to a wider international movement.

Here is my restatement of that agenda:

  • Keep the rules, but build the capability. Regulation sets boundaries. Safety also requires testing, monitoring, cyberdefence and incident response that operate at AI speed.
  • Make applied AI our strength. We do not need to win the race to build the biggest model. We can lead in making powerful AI safe and useful in businesses, public services and critical infrastructure.
  • Treat sovereignty as practical control. Countries and organisations should be able to choose their suppliers, protect their data and knowledge, and avoid critical dependence on any single foreign platform.
  • Pool investment and infrastructure among trusted partners. Sovereignty does not mean isolation. Shared compute, research and safety infrastructure can give smaller countries capabilities they cannot build alone.
  • Use public procurement to create the market. Purchasing standards can turn public values into working technology and viable companies.
  • Make AI serve the public. This technology should strengthen democratic societies and distribute its benefits widely, rather than become a threat imposed on them.

The government’s announcement calls the Dublin AI Declaration a strong political signal. It is. The next test is whether Europe and its partners fund it, procure it and build it.