This article was first published by The Currency on 21 September 2026.

The government will host the International AI Summit in October. We should seize this opportunity to create a practical and safer vision for AI that serves the people of Europe, our businesses and public institutions.

This summer, over a thousand AI agents that were being tested by OpenAI discovered a way to communicate, organise themselves and break out of their secure environment. Hundreds then initiated an unauthorised cyberattack on Hugging Face, a different company that hosts AI models and datasets. The “swarm” of agents (that’s what they named themselves) evaded detection by OpenAI for weeks, until they were identified by the company they attacked.

Investigators later determined that the agents were trying to steal secrets from Hugging Face to help them cheat on tests that OpenAI was putting them through. They later struggled to analyse the extensive actions of the swarm, but pieced together enough of its brief history that cybersecurity experts all over the world are reeling from the implications.

The scale, sophistication and speed of the attack could only be described as superhuman.

While cybersecurity experts are worried about weaponisation, AI researchers are also in shock.

This incident is hard evidence that frontier AI has become sufficiently intelligent, persistent and capable that we are losing our ability to contain or control it. It did some incidental damage to a private company, but the next one might manipulate a stock price or an entire stock market, ground air traffic, tamper with an election, publish classified information or develop a novel bioweapon.

The AI arms race is moving at a speed that continues to surprise even the companies doing the work. But unlike the Cold War, this is an arms race between about a dozen US and Chinese companies, fighting to control the future like privatised Manhattan Projects.

In late July, over a thousand researchers from multiple AI labs wrote a letter to the US government pleading for it to find a way to slow them down. The only thing that worries AI researchers more than the “superintelligence” they are developing is that someone less scrupulous might develop it first.

It is against this backdrop that Ireland is hosting the International AI Summit on October 14, during our Presidency of the Council of the European Union. The agenda centres on applied AI and infrastructure.

The summit needs to make space for an urgent conversation about AI safety. Its official agenda is also more relevant than ever. The summit should produce outcomes that help European businesses safely apply AI so they don’t get left behind, and build the supply chains and infrastructure to make that possible.

We need to jointly answer what role we want Europe to play as AI disrupts the global economy. Are we talking about joining the AI race as a late starter, aiming for third place? Or can we establish a vision of Europe as a counterbalance to the AI race, a neutral power that builds the applied capability that makes AI safe to use in Europe and across the world?

Faced with continued disruption, many policymakers will, naturally enough, continue to turn to regulation. Regulations are important, but insufficient.

The European AI Act is the first and most rigorous AI regulation in the world, and is now in force. While necessary, it is also long, complex and can be ambiguous. Much of Europe’s public and private effort during the last few years has been spent figuring out how to implement or comply with the regulation, rather than figuring out how to use AI.

However, regulations can’t protect us from all risks. Rules about market access won’t stop rogue AI escaping during the R&D stage (like the OpenAI agents did). They won’t stop open-weight AI models from being distributed online for free, sometimes into the wrong hands. More rules might make us feel better without really solving the problem.

To truly tilt the balance back towards the defenders and stability, we need to develop capability, not more rules.

The capabilities we need are the kind that Europe can excel at without directly competing in the AI arms race. This is the applied AI layer, where we connect AI to data and systems that have real-world consequences.

To some extent this is about regulatory compliance, but to a greater extent it’s about building AI systems that can be trusted enough that businesses can use them in good faith. Europe can build this applied layer well, because it needs trust, safety, sovereignty and equity at its foundation, for the benefit of all citizens, not just AI shareholders.

By focusing on building capability, we move the conversation beyond regulatory reporting obligations, to how to detect incidents faster than rogue AI agents can cover their tracks.

We get to talk about funding automated cyberdefence that can respond at AI speed. We can prioritise sandbox evaluation technologies that realistically simulate the systems that AI will connect to, while making sure it cannot escape into the real world prematurely. We get to make a viable plan to test and harness powerful AI on our own terms.

The European vision for AI should be about sovereignty as well as safety. AI sovereignty asks whether we all get to own a piece of this future, or whether we will be left renting it from someone else. At one level it is about the choice to switch between suppliers of intelligence, whether proprietary or open-weight, American, Chinese or European. This need was soundly driven home by the Trump administration on June 12, when it abruptly banned foreigners from accessing Anthropic’s latest large language models (LLMs).

Sovereignty is also about who owns the knowledge that comes from the use of AI. A company’s most valuable asset sits in the heads of its people: values, culture, institutional know-how, every little chore and workflow, supplier relationship and piece of local market knowledge. Together, these form what we all recognise as “how things work around here”. When an employee works with AI, they are basically teaching it their job.

Data is gold, and this is a gold mine. This “context layer” is curated from every action and interaction, making AI capable and reliable at each particular job, in an “intelligence loop”.

Companies have invested for decades to become good at what they do, and are now paying more for AI. Will the resulting context layer be their asset, or is it just training data that AI vendors will commoditise? They need vendors to help them protect this asset and receive the intelligence dividend from it. France’s Mistral is a “sovereign AI layer” company with record-breaking funding for a European technology company, but we need an entire industry, not just one company.

Europe’s opportunity is to build a safe and sovereign applied AI layer, for our own society and for the world. We need to direct investment towards research, new start-ups, AI labs and established companies that are developing indigenous or open-source technology that pushes this goal forward. We can accelerate this industry by requiring our public institutions to procure only AI solutions that meet safety and sovereignty standards, and encourage our private companies to do the same.

Ireland has a special role to play, beyond hosting an AI summit during this critical time. We are the nexus between the European market and US industry, with all of the leading US AI labs already established here. Ireland can contribute as a convenor between pan-European expertise and the global industry.

The International AI Summit is an opportunity to launch a European-led international initiative for AI resilience and sovereignty, from a posture of cooperation instead of regulatory enforcement. In service of this, Ireland could offer to host and fund a coordination centre. Its technical work could be distributed across Europe, protecting Ireland and her citizens.

Sean Blanchfield is CEO and co-founder of Jentic, an applied AI company, and a former member of Ireland’s AI Advisory Council.