Everyone wants to control AI. But nobody really knows how

The debate on how to regulate artificial intelligence has exploded. Pope Leo XIV, in his first encyclical, called for AI to be placed at the service of people, work and the common good, and not to become a tool for the concentration of power. More than two hundred economists, including Nobel laureates, have signed the ‘We Must Act Now’ appeal, calling for institutions capable of managing its effects on employment and income distribution. The main merit of that appeal, beyond the individual proposals, is perhaps another: it has made it clear that by now almost everyone wants to regulate artificial intelligence, but no one really knows how to do so. One might call for more public oversight, more competition, greater corporate accountability, and more redistribution. The problem is determining which tools actually work without stifling a technology that can also yield enormous benefits.

Lina Khan, former chair of the US Federal Trade Commission, proposes using antitrust legislation to prevent a small number of companies from jointly controlling models, data, chips, cloud infrastructure and distribution channels. In the background are the thousands of lawsuits against Meta and other social media platforms, accused of designing Instagram and Facebook to be addictive, particularly amongst minors. These are not lawsuits concerning AI in the strict sense, but they serve as a reminder that algorithms can cause collective harm long before the authorities are able to understand and regulate them.

OpenAI is also said to have proposed that the public should share in the economic benefits of artificial intelligence by transferring 5 per cent of its capital to the US government and calling on other large companies to do the same.

The proposals are therefore very diverse: ethical principles, competition law, civil liability, public shareholdings, and profit redistribution. Regulation that is too weak risks coming into force only once markets are already concentrated and the damage has been done. Regulation that is too rigid, on the other hand, may stifle useful applications, favour precisely those dominant firms that have the most resources to adapt, and leave behind the countries that adopt it.

This is where the European situation becomes interesting. Europe is notoriously accused of being unable to foster the growth of new sectors, but only of regulating them. But in this case, is this a good or a bad thing? From 2 August 2026, European regulation will have a direct impact on the day-to-day operations of businesses and organisations using artificial intelligence systems. It does not apply to those who build the large models (which, as is well known, are not European), but to those who use them in their own processes, in their dealings with customers and employees, and in the production of content. Businesses are required to meet obligations regarding transparency, training, monitoring the use of the systems and identifying responsibilities.

The question is no longer simply whether artificial intelligence should be regulated. It is whether such regulation is useful enough to justify the costs and effort involved in compliance for businesses, or whether it will end up becoming pointless and harmful bureaucracy. This is the real test for Europe.

So far, in our view, the answer is positive. From February 2025, businesses and organisations must take steps to ensure that those using AI tools have sufficient skills to do so responsibly. From 2 August, Article 50 of the AI Act will then take centre stage, requiring that, in certain circumstances, interaction with artificial systems and the synthetic or manipulated nature of certain content be made recognisable.

In practice, a local authority or a company using a chatbot must make it clear when a customer is speaking to a machine; an estate agent publishing artificially generated images must provide the correct disclosure; a company using AI in recruitment must know who is in control of the tool and who is accountable for the decisions. The principle is simple: AI cannot be used invisibly and without identifiable accountability.

And it is perhaps a good thing that the most important and onerous obligations, particularly for high-risk systems, have been postponed. When AI is involved in decisions that directly affect people’s lives, regulation is certainly necessary. But here, the risk of getting it wrong is much greater: premature regulations can stifle technologies that are still evolving, create enormous costs and favour the largest operators.

Europe therefore faces a crucial challenge: to demonstrate that artificial intelligence can be regulated without stifling it. So far, the AI Act appears to be moving in the right direction: transparency, expertise and accountability are manageable costs that yield tangible benefits. When it comes to the more onerous regulations, however, it would be better to have allowed more time. Because today, everyone wants to regulate artificial intelligence. The problem is that no one yet fully knows how to go about it.