AI needs its own IAEA
Liisa Pakosta, Minister of Justice and Digital Affairs of Estonia
The development of AI and the services and autonomous systems built on it has reached a point where it's no longer enough to focus on building one nice, useful domestic e-service after another. It's time to reach a global agreement on the shared, ethical use of AI infrastructure — in internationally used e-services, and in a global system for threat warnings and risk mitigation.
Nuclear energy became a boon for humanity rather than a catastrophic threat because we managed to agree, quickly, on shared rules and to build an international institution to enforce them: the International Atomic Energy Agency. With artificial intelligence, we face the same choice — except this time we need to decide much faster.
AI is a borderless phenomenon. National rules alone are no longer enough — just as national or even European law alone would never have been enough to regulate nuclear energy. If we want to unlock AI's potential and contain its risks, we need global standards, and an institution to back them up: an IAEA for artificial intelligence.
What would such an organization look like, and what would it do?
A global computing power solidarity mechanism. To narrow the digital divide, countries could build a shared resource-pooling system that guarantees access to critical models — independent of closed ecosystems — for developing nations too, in areas such as climate-change mitigation, pandemic prevention or education. That would drive genuine digital equality, rather than merely a declared intention to pursue it.
A global standard for labeling AI-generated content. Limiting the effort to Europe and the United States won't cut it. We need a unified system for tagging AI-generated content — a digital fingerprint — a mandatory global watermarking regime for the largest developers, recognized by platforms and national digital services worldwide. That is a precondition for protecting democracies and a clean information space.
An international early-warning network. Much like the international aviation-safety system, we need a real-time reporting network to share information on AI anomalies and security vulnerabilities. Only that will let us head off global cyber crises and keep critical services running without interruption.
None of this is theoretical. Estonia is trying to lead by example. Our Eesti.ai initiative, run jointly with several ministries, is already putting AI to work in concrete fields; AI Leap is bringing the tools and the critical-thinking skills straight into classrooms; and we keep adding safeguards to our own laws as we go. But just as with nuclear energy, national or even regional solutions alone won't do — we need a global framework to stitch these separate initiatives together.
At the Tallinn Digital Summit this November, there's little point dwelling on AI's theoretical benefits or the fears surrounding it. The real question is how to build a governance model people can actually trust — and that requires international cooperation on the same scale that once made nuclear governance work.
That is the message Estonia wants to send to everyone coming to Tallinn: trust and human rights are not constraints on innovation — they are the foundation that gives innovation real value. A state built on AI can only be citizen-centric, democratically accountable and grounded in the rule of law — and in the long run, those principles are the basis of competitiveness, not an obstacle to it.