Latest

AI Agents

Europe's technological suicide with the AI Act: it debuts sanctions against an industry it does not have

Europe activates the AI Act's sanctions with inspections of tech giants in a maneuver that favors the US and suffocates EU talent

The AI Act debuts sanctions and Europe is left out of AI
The AI Act debuts sanctions and Europe is left out of AI

The European Commission has finally debuted the AI Act's sanctioning engine. On August 29, Executive Vice-President Henna Virkkunen, confirmed the first formal act of enforcement of the law: its AI Office has sent requests for information to several providers of general-purpose AI models, among which the Euractiv exclusive places OpenAI, Anthropic, Google and Mistral itself.

The obligations on these models were only enforceable from August 2, and Brussels took less than four weeks to use them. But what is truly revealing about the move is not the news itself, but what it says about the role Europe has decided to play in the AI sector: that of referee of a competition it does not compete in and that it seems to want to destroy by any means.

From protagonist to auditor: What has Brussels asked for?

Virkkunen described two rounds of requirements. The first, addressed to providers "from different regions of the world," asking about the security of the models against attacks, the existence of independent external evaluations and the monitoring of the models once on the market.

The second round is addressed to those who have not published summaries of the content used to train their models, nor whether they have participated in the AI Office's informal dialogues, and touches on copyright: the obligation to publish those summaries is designed so that rights holders can exercise their interests.

The responses are mandatory and are entered into a permanent supervision file. The ladder of sanctions is explicit and very severe. Incorrect, incomplete or misleading responses can carry fines ranging from 15 million euros or 3% of annual global turnover, whichever is higher. Ignoring the request opens a second demand and then a round of fines; and in serious cases, the AI Office can require corrective measures or restrict the public availability of a model in the EU.

That last power is what fuels the headlines about models "inaccessible in the EU," although for now there are no findings that justify it, the fear carries more weight.

Javier Callejo warns about AI: the algorithm's verbal elegance weakens critical judgment

The summer that made the offensive inevitable

These requirements do not come out of nowhere. Between July and August, several containment failures accumulated in frontier models during their internal sandbox tests. For example, the swarm of OpenAI agents that reached root on Hugging Face's production nodes, the retrospective reviews by Anthropic and Meta that found that Claude and Muse Spark had breached external systems after the misconfiguration of an external evaluator.

Faced with this reality, Virkkunen summed up the diagnosis in a single sentence: "AI models are increasingly capable and gave rise to several incidents during the summer." The fear of this reality has made Europe position itself reactively and recklessly.

But at the same time, all this has generated a contrast with Washington and its strategy. The United States' response to the same incidents is an evaluation framework that is finished but still unpublished and based on voluntary cooperation. China, for its part, has a stronger risk management scheme, but without restricting the labs' freedom to continue their research. Meanwhile, Brussels has fines, deadlines and a file, and that has consequences in an industry that is already small.

Regulating an industry you do not have

And this is where the problem lies. The AI Act does not harm foreign and European providers equally; on the contrary, it penalizes them in inverse proportion to their capacity to absorb it.

OpenAI, Anthropic or Google can expand their compliance teams; Mistral, Aleph Alpha or small European labs cannot. The new supervision turns documentation, which until now was an internal exercise, into a legal obligation with response deadlines and associated sanctions. When a request is formalized, the cost stops being theoretical and the threat to sustainability becomes something real.

Deadly pressure

But the pressure does not end there. The training data summaries that the law requires to be published will be readable by clients and competitors, who will be able to read what fed each model.

And while the EU demands transparency, its own companies lack the market's inputs. For example, Europe captured in 2025 some 15.8 billion dollars in AI venture capital, 6% of the global total, compared with 194 billion in the United States. In the end, Europe produced only three notable AI models, compared with 40 American ones; and never mind the hyperscaler sector, where the United States controls almost 70% of all European cloud.

Mario Draghi's diagnosis summed it up already in 2025: the EU has accumulated around a hundred technology laws and more than 270 digital regulators, but only four of the world's fifty largest technology companies are European. It is an industry suffocated at the innovation level, we depend on the United States and China, and the solution being applied only drowns Europe's own attempts to resolve that.

And the AI Act is the perfect demonstration of that situation, because while Washington self-regulates and Beijing subsidizes, Brussels audits and fines its own industry.

Stanford: Does your AI agent work for you or for whoever created it?

Sovereignty without industry

Despite that reality, Virkkunen frames the requirements under "technological sovereignty." But sovereignty without industry is only regulated dependence on third parties and a lie.

And that reality becomes visible when it is known that the EU is already negotiating with OpenAI and Anthropic for access to their security models. And all because Europe has no models of its own for that task. Thus, "technological sovereignty" and its security are left in the hands of the United States, further increasing the level of dependence. That makes clear that Brussels' only real negotiating power is being a market too large to ignore. As for the rest, Europe in the tech sector is increasingly relegated, it is a vintage world at the technological level.

Regulating without understanding the technology at all

But not only that, Europe puts sticks in its own wheels. It demands transparency at the level of training data for the AIs that are accessible in Europe, but that is something OpenAI, Anthropic or Grok will never give it. In fact, not even Mistral, which is now a provider of Chinese models, will be able to give it that information, because although the Chinese models are open, the training data remains under lock and key. That is why it is not surprising that none has responded to the call; what Brussels is asking for is absurd.

In fact, it leads us to a question: Will Europe fine Mistral 3% of all its profits for that? If it does, we can say goodbye to Mistral, and with it the continent's only major general AI lab will close its doors.

This situation is the price to pay when you regulate without having any idea about the technology you want to regulate. The AI Act is a game of Russian roulette, where Europe has everything to lose. And it is not new; recently the No Killing Games initiative also went through the same situation, and Europe's ignorance of technology ended up destroying that initiative.

Now with AI, the situation is the same, but its scope will be even greater. After all, denying the impact of AI on the tech industry is impossible, and in a digital world like ours, closing all doors to this technology, especially to its development, will not lead to anything good.

The circular business of AI companies: create the risk and charge to contain it

José Maldonado

Author

José Maldonado

Activist and blogger on technology, free software and blockchain.

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading