“Rules exist to protect people’s rights, not simply for the sake of regulation.”
“Europe must build its own digital sovereignty without giving up innovation or the values that define it.”
“If the environmental impact of artificial intelligence were transparent, citizens, businesses and public institutions would make very different decisions.”
Nicoletta Rangone is one of Europe’s leading experts in administrative law, regulatory quality and the governance of artificial intelligence. She is Full Professor of Administrative Law at LUMSA University in Rome, where she directs the Law degree programme and is a member of the University’s Centre for Research and Internationalisation. She also teaches in the Erasmus Mundus European Master in Law and Economics (EMLE) and previously held a Jean Monnet Chair dedicated to the European approach to Better Regulation.
She recently became Director of the new Sustainable Artificial Intelligence for Regulation (SUSTAIN-AI-REG) Centre of Excellence, a project funded by the European Commission that will, over the next three years, examine how to reconcile the development of artificial intelligence with the protection of fundamental rights and the environment. She is also a member of Italy’s Regulatory Impact Assessment Unit (NUVIR) and has advised organisations including the OECD, the World Bank, the Italian Competition Authority, and the National Anti-Corruption Authority. Rangone believes that Europe should neither imitate the American model nor the Chinese one, but instead develop its own path based on fundamental rights, transparency and technological sovereignty.
You have just launched a new Jean Monnet Centre of Excellence dedicated to sustainable artificial intelligence. What is its objective?
This is a three-year project that has only just begun and builds upon a line of research we have been developing over the past several years through the European Commission’s Jean Monnet programmes. Our original work focused on Better Regulation, and quite naturally that field eventually intersected with artificial intelligence. AI has the potential to become an extraordinarily valuable tool for helping public administrations draft better legislation, conduct ex-ante impact assessments and manage large-scale public participation processes. It can significantly amplify these regulatory techniques, but it also introduces important risks, particularly in terms of inclusion, transparency and the protection of fundamental rights. That is why we decided to broaden our research towards an issue that has so far received much less attention: the environmental sustainability of artificial intelligence.
Why do you believe AI’s environmental impact remains one of the major unresolved issues?
When I first began studying artificial intelligence, I realised that alongside its legal and social risks there was another dimension that was almost entirely absent from the public debate: its environmental costs. These costs are very real and highly significant, yet they often remain invisible because they are treated as externalities that are borne neither by developers nor by users. We are talking about electricity consumption, the water required to cool data centres, the production of semiconductors and the entire lifecycle of digital devices. The real question is who should bear these costs and how they should be incorporated into the regulatory framework.
The AI Act barely addresses this issue. Was that a missed opportunity?
The European Artificial Intelligence Act is an extraordinarily important piece of legislation, and Europe can rightly be proud of having become the first major jurisdiction in the world to establish a comprehensive legal framework for artificial intelligence. Nevertheless, it also has clear limitations, one of the most significant being the very limited attention it pays to AI’s environmental impact. I believe this will become one of the defining issues of the coming years, because sustainability cannot remain a secondary consideration in technological development.
In recent months, many people have argued that Europe regulates too much and that this hampers innovation. Do you agree?
I don’t think the debate can be reduced to such simplistic terms. Rules do not exist because regulators want to justify their existence; they exist to protect people’s rights. In that respect, Europe is fortunate because its AI Act is clearly rights-driven, placing fundamental rights at the very centre of the regulatory framework. Other parts of the world prioritise innovation above almost everything else. Europe, by contrast, seeks to reconcile both objectives, and in my view that is precisely its greatest strength.
So the real issue is not choosing between rights and innovation?
Exactly. It is a false dilemma. We cannot protect rights to the point of preventing any form of innovation, but neither can we sacrifice them in the name of technological progress. The challenge is to move forward on both fronts simultaneously, maintaining a constant balance between protecting people and fostering innovation.
However, you have identified some shortcomings in the way Europe legislates.
Yes. I believe the objectives of the AI Act are excellent, but its greatest weakness is its extraordinary complexity. The Regulation alone contains 113 articles, 180 recitals and 13 annexes. On top of that, there are delegated acts, implementing acts, harmonised standards, codes of practice and guidelines that are still being drafted. The result is an extremely complex legal architecture that creates uncertainty for both businesses and public administrations. Europe certainly needs to simplify this system, but simplification must not mean removing safeguards. Simplification should mean reducing unnecessary administrative burdens while preserving the protection of fundamental rights. Otherwise, we would no longer be talking about simplification, but deregulation.
The European Commission has already begun revising legislation that has not yet been fully implemented. Do you see risks in that?
The greatest risk is legal uncertainty. Following the Draghi Report, the European Commission introduced several omnibus packages aimed at revising legislation such as the AI Act and the General Data Protection Regulation. Some amendments may prove beneficial because they remove unnecessary burdens, but others represent a step backwards. The problem is that legal obligations are being changed before businesses and public authorities have even had the opportunity to implement them. Complying with European legislation already involves significant costs. If the rules keep changing, the cost of compliance rises further, and so does the temptation to delay implementation—or not to comply at all.
You have also criticised the process through which these reforms are being adopted.
Yes, because we cannot solve regulatory complexity by weakening democratic decision-making. Some of these legislative packages were prepared through very limited consultations aimed primarily at experts, without sufficient public participation or proper ex-ante impact assessments. Europe has long distinguished itself through participatory mechanisms such as the Have Your Say platform, which enables citizens to contribute to the design of public policies. Abandoning these safeguards when reforming legislation such as the AI Act or the GDPR means moving forward almost blindly. The solution is not to simplify the legislative process itself, but to produce legislation that is clearer and less burdensome.
After visiting Barcelona, Madrid and Valencia, you praised the innovative approach of Spanish public administrations. Were you surprised?
I was genuinely impressed to meet public officials who were so open to innovation. The message from those cities was essentially: bring us your ideas and we will make our urban data available. What struck me most was that all of this takes place under exactly the same GDPR that applies in Italy. The difference therefore does not lie in European legislation, which is identical across Member States, but in the way national institutions interpret it. In Spain—at least in the cases I had the opportunity to study—data protection has not prevented innovation. In Italy, however, some interpretations have been considerably more restrictive.
Italy was the first European country to temporarily suspend ChatGPT. Was that an example of this restrictive interpretation?
Yes. The decision was based on concerns regarding the legal basis for processing the data used to train the models, the accuracy of that data and the absence of adequate age-verification mechanisms. I do not question the importance of protecting privacy, but the same legislation can be interpreted in very different ways. Legal interpretation must be both evolutionary and proportionate. Protecting fundamental rights is essential, but it should not override other legitimate objectives, such as innovation in public administration. When public authorities use these technologies, their purpose should always be to serve the public interest and improve services for citizens.
France appears to have made greater progress than Italy in digital sovereignty. How do you explain that difference?
France has always had a strong tradition of supporting its own national champions. Today, that translates into a concrete strategy for digital sovereignty: public investment in cloud infrastructure, support for companies such as Mistral and restrictions on the use of certain foreign platforms in sensitive sectors. These alternatives may not offer all the features available from the largest American technology companies, but that is not really the point. Public administrations must also consider where their data are stored, which legal system protects those data and the level of technological dependence they are creating.
Several Italian universities provide ChatGPT to students and faculty members. Why do you consider that problematic?
I am not saying these tools should not be used. What I am saying is that public institutions should be much more aware of the message they send when they deploy a specific platform on a large scale. When a university provides a product to its entire academic community, it implicitly suggests that the product is both safe and trustworthy. Yet universities handle personal data, sensitive information, research results and other materials that should not be shared without appropriate safeguards. Furthermore, once an organisation adopts a platform, moving away from it later becomes extremely difficult. That dependency is also part of the broader debate on digital sovereignty.
Does speaking about European digital sovereignty amount to protectionism?
Not necessarily. Digital sovereignty is not about isolationism or technological nationalism. Rather, it is about ensuring that the essential technologies we rely on are consistent with our legal principles and democratic values. Europe has built a model based on fundamental rights, data protection and respect for human dignity. That sensitivity is not shared equally across all legal systems. If we claim that our approach is rights-driven, then we must remain consistent not only in our legislation but also in our technological choices and public procurement policies.
Is the European rights-based approach truly compatible with innovation?
It has to be. The challenge is always to strike the right balance. The AI Act, for example, prohibits the use of artificial intelligence to infer people’s emotions in certain contexts, such as education. That prohibition may limit some personalised learning applications based on analysing students’ emotional responses. At the same time, however, it may prevent a young person from being permanently labelled as a slow learner, emotionally unstable or exceptionally gifted. Europe may decide not to pursue certain technological applications, but it will encourage others that are more respectful of human dignity. Innovation does not necessarily mean adopting every technology that becomes available.
It is striking that artificial intelligence has not become a deeply polarising political issue in Italy.
Yes, it is probably one of the few subjects that has not yet become politically polarised. In Italy, almost everything tends to generate political divisions, but artificial intelligence has largely escaped that dynamic. In principle, that is positive because it prevents such a complex issue from being reduced to ideological slogans. What concerns me more is that the subject is still not receiving enough public attention. There is still insufficient awareness of AI’s broader implications, particularly regarding its environmental costs and its impact on public policy.
You often argue that sustainability is not only an environmental issue but also a matter of governance.
Exactly. When we discuss the costs of artificial intelligence, we often think in very abstract terms. Yet many of its consequences are already highly tangible. In some countries, for example, energy companies have stopped signing supply agreements with local communities because the demand generated by data centres consumes a significant share of the available electricity capacity. We also know that conversational AI systems require considerably more energy than many other types of AI models. All of this should prompt us to ask when these systems are genuinely necessary and when more efficient alternatives may be available.
This debate directly affects public administrations as well.
Without question. Not every public administration needs large general-purpose language models. In many situations, Small Language Models (SLMs) may be far more appropriate because they consume fewer resources, are easier to supervise and allow administrations to reuse knowledge already developed elsewhere. Public administrations share many common functions—inspections, risk management, administrative procedures and regulatory oversight. If one administration has already trained an effective AI system, it makes perfect sense for another administration to reuse it rather than starting again from scratch. Doing so reduces financial costs, lowers energy consumption and also makes it easier to ensure transparency and traceability.
Should software reuse and open-source solutions become European priorities?
I believe they should. In Italy, the Agency for Digital Italy already recommends giving preference to open solutions, but that can only work if transparency also exists. We need to know what each administration has developed, which tools are available and what results they have achieved. Sharing that knowledge is essential. European cities face many of the same challenges, so there is little sense in each of them reinventing identical solutions. The objective is not simply to copy one another, but to learn from successful experiences and adapt them to different contexts.
One of your main proposals is to establish common indicators for measuring AI’s environmental impact.
This is probably the greatest challenge we face. At present, we do not have a shared methodology for measuring the environmental footprint of artificial intelligence systems. We still lack agreement on what should actually be measured, where an AI system’s lifecycle begins and where it ends. We talk about data centres, but we must also consider semiconductors, user devices and their eventual recycling. Every company currently performs its own calculations, making meaningful comparisons impossible. Without common metrics, genuine transparency simply cannot exist.
And without transparency, informed choices become impossible.
Exactly. If citizens, businesses and public authorities knew the real environmental impact of different AI systems, they would make very different decisions. Transparency has a transformative effect because it encourages more efficient solutions while creating reputational incentives for companies to improve their environmental performance. Today, however, the situation remains extremely opaque. Major technology platforms treat this information as commercially sensitive because it represents a competitive advantage, making objective comparisons extraordinarily difficult.
Does the DeepSeek case demonstrate that it is possible to innovate by reusing existing knowledge?
It is one of the most interesting lessons to emerge from the DeepSeek case. Its development shows that innovation does not necessarily require starting from scratch every time. It is possible to significantly reduce the resources required by building upon existing models, knowledge, architectures and infrastructures. This reinforces the idea that reuse can itself become a principle of digital sustainability. Just as the circular economy promotes the recovery and reuse of materials and components, artificial intelligence should, wherever possible, encourage the reuse of models and computational resources, avoiding unnecessary duplication. The concept of circularity should therefore not apply only to physical objects but also to digital systems.
At the same time, we must also consider the recent allegations made by Anthropic against DeepSeek, which illustrate how model reuse and knowledge distillation raise delicate issues concerning access conditions, licensing, transparency and the protection of the investments made by the original developers. Promoting reuse must therefore go hand in hand with rules capable of balancing sustainability, innovation and the protection of the rights and legitimate interests involved.
When you speak about technological sovereignty, are you referring to national sovereignty or European sovereignty?
Always European sovereignty. I consider myself profoundly European. I belong to the first generation that experienced the Erasmus programme, and that profoundly shaped my understanding of Europe. Whenever I travel to Spain, France or any other European country, I feel at home because, despite our cultural and linguistic differences, we share a very deep set of common values.
That is precisely why I believe Europe must build its own technological capacity—not in order to compete from a nationalist perspective, but to preserve a model of society based on fundamental rights, democracy and the rule of law.
Pope Leo XIV has recently published the encyclical Magnifica Humanitas, which also addresses the environmental impact of artificial intelligence. What is your assessment of it?
What struck me most was how explicitly the Pope addressed this issue. Environmental concerns had already appeared in previous encyclicals, but this time he speaks directly about water consumption, energy use and the impact that artificial intelligence may have on future generations.
It is significant that these reflections are now also coming from spiritual leaders and international organisations such as the United Nations. This shows that the debate has moved well beyond the purely technological sphere. Artificial intelligence is now a matter of collective responsibility, and we must be prepared to face not only its benefits but also the costs it entails. Only then will we be able to unlock its full potential without compromising the future.