The laws hedge in these new beginnings and guarantee the preexistence of a common world, the permanence of a continuity that transcends the individual life span of each generation, and in which each single man in his mortality can hope to leave a trace of permanence behind him.
H. Arendt, Thinking without a Banister: Essays in Understanding 1953–1975, p 46.
I. Introduction
This article proposes the novel concept of justice-centred artificial intelligence (AI) through a lens, Hannah Arendt’s political thought, and a use case, the widespread use of AI in energy systems. Inspired by existing research and Arendt’s humanistic perspective, this article reclaims the principles by which we act and the criteria by which we judge and conduct our livesFootnote 1 to establish a digital constitutionalism integrative of ecological elements.Footnote 2 Such an interdisciplinary perspective aims to investigate existing regulatory gaps at the intersection of AI and energy, which the International Energy Agency (IEA) has dubbed the new ‘power couple’.Footnote 3
This article proceeds in four main sections. Section II presents the methodology. Section III defines the gap between the democratic ideal of collective deliberation and the reality of individuals behaving as energy consumers. Section IV explores the role of international and regional law in filling the gap by offering a baseline of three principles, as explored in Section V, for justice-centred AI, particularly in energy matters. Finding these principles in need of institutionalisation, Section VI proposes three methods to embed the constitutive principles of justice-centred AI through both centralised and decentralised means.
This article adopts a European digital constitutionalism perspective, situating its analysis within the framework of EU constitutional law, including the Charter of Fundamental Rights of the European Union and the evolving body of secondary legislation governing AI, most notably the AI Act. While informed by international human rights and environmental law, the primary geographic and jurisdictional scope of this contribution is the European Union.
From a normative standpoint, the article operates at the intersection of de lege lata and de lege ferenda. On the one hand, it interprets existing EU legal materials, including fundamental rights, general principles of EU law, and regulatory instruments. On the other hand, it allows for a normative reconstruction of these materials through the articulation of justice-oriented principles intended to guide the development and deployment of AI systems beyond the current regulatory framework. These principles therefore should be read not as purely descriptive of the law as it stands or as abstract philosophical ideals but as constitutional proposals embedded within the EU legal order, aimed at addressing identified normative gaps.
This focus responds to the absence of a coherent set of guiding principles in the EU Artificial Intelligence Act (AI Act), a structural shortcoming that undermines also the possibility to shape a regulatory model centred on human rights.Footnote 4 The European Parliament’s introduction of such principles, largely by replicating those developed by the High-Level Expert Group on AI, did not substantially improve the AI Act on this issue.Footnote 5 The resulting framework risks functioning as a vademecum of pre-existing values, often redundant with established EU law (eg data protection, non-discrimination) or too vague to effectively guide AI development and deployment.Footnote 6 In this sense, the absence of clear guiding principles in the AI Act is not merely a legislative gap but a constitutional one, also vis-à-vis more principled pieces of EU digital legislation such as the General Data Protection Regulation (GDPR).Footnote 7
This approach aligns with the literature on European digital constitutionalism, which explores in particular the protection of fundamental rights and the regulation of private digital actors.Footnote 8 In particular, the EU is currently transitioning from digital liberalism, prioritising economic innovation, to digital constitutionalism, which explores (i) the horizontal application of human rights and (ii) the alignment of digital law with existing regulatory frameworks.Footnote 9
A necessary limitation of the proposed principles concerns their jurisdictional and enforcement reach. The proposal is inspired by the EU AI Act’s extraterritorial logic (Art 2), which extends its scope beyond formal market placement or deployment. In particular, the Regulation anticipates scenarios in which it governs AI systems that are operated outside the Union and yet the resulting outputs are used within the EU, thereby preventing regulatory circumvention and ensuring effective protection of fundamental rights. The practical enforceability of such extraterritorial obligations, however, remains largely untested. This is particularly relevant for the present proposal’s reliance on duties of diligentia diligentis (the diligence of a reasonably prudent person) imposed on AI providers and deployers, which presupposes a degree of regulatory reach that may not fully materialise, especially in transatlantic contexts, given widely different regulatory approaches and risks of regulatory arbitrage (see Section IV.B).
Within this context, the reference to Hannah Arendt serves a methodological function. Arendt’s idea of a ‘right to have rights’ and her call for a legal order capable of addressing humanity as a whole, as later explored, are invoked to interpret the constitutional fabric of current AI regulatory efforts in EU law. In this sense, Arendt’s thought is mobilised to illuminate the tension between universality and situated legal orders, as well as to support the articulation of principles that, while grounded in EU law, aspire to broader normative relevance. Accordingly, this article understands the EU as more than a regulatory actor. At the present juncture, the EU rather seems a constitutional laboratory in which the principles governing AI can be developed, tested, and potentially projected beyond its borders.
II. Background, assumptions, and limitations
Offering an Arendtian take on AI and its place in the world of international law is not as baffling as it may initially appear. Notably, Arendt was a thinker endowed with a profound historical consciousness and a pioneer of many of the key themes of the 21st century: progress, the crisis of modern science, the rise of the bureaucrat and mass societies, as well as the opposition between culture and technology. Moreover, AI was first conceptualised during Arendt’s life, becoming as much a product of the 20th century as it is of the 21st century.
The term ‘artificial intelligence’ was coined in 1956 by information theorist Claude Shannon during a conference at Dartmouth College that was sponsored by the US Defense Advanced Research Projects Agency.Footnote 10 However, Alan Turing’s pivotal paper in 1950, published in Mind, already suggested new avenues of inquiry: can a machine think? Can a machine be linguistically indistinguishable from a human? Turing also introduced new methods of inquiry, such as the Turing Test, a tool to evaluate a machine’s capability to demonstrate intelligent behaviour akin to that of a human in an ‘imitation game’ of sorts.Footnote 11 The Turing Test posed a philosophical inquiry into how minds work, a theme also dear to Arendt.
Arendt’s engagement with questions that later resonated with debates on AI must be situated within her broader philosophical project. While AI was still in its infancy in 1958, Arendt identified the change in the constellation and mutual relationship of human capabilities from animal laborans to homo faber up to individuals capable of action (homo politicus), with all their self-created risks in an increasingly technological world.Footnote 12
Building on this insight, Arendt articulated a framework of human capacities that structures the way we relate to nature, the world, and one another, which can be classified into three capacities: labour—dictated by natural needs; work—poised to create a safe space for existence; and action—the coming together as equals in a community. Further, the ability to act and speak differentiates human beings from other animals or the operations of machines.Footnote 13 Even more importantly, Arendt’s phenomenological deconstruction reveals how the machine world has become a substitute for the real world, even though such a ‘pseudo-world’ cannot fulfil the most important task of human action, which is to provide mortals with a more permanent and stable dwelling than themselves.Footnote 14 To enable such a dwelling, Arendt complemented her theory of action among individuals (vita activa) with a theory of activity within individuals (vita contemplativa), namely the life of the mind.Footnote 15 In Arendt, the law’s place is within action: it serves as the walls for the political life of citizens, endowing power with institutional stability and increasing the scope of legitimate action.Footnote 16
Arendt emphasised that, for the first time in history, the human capacity for action had begun to dominate all other capacities—the capacities of homo faber and human animal laborans, and the capacity for contemplation. Indeed, since the 20th century, the human capacity for action has been epitomised by technology, which has arisen as the meeting ground of history and nature.Footnote 17 To be sure, technology has proven able to act not only over but into nature as we used to act in human affairs.Footnote 18 The technological world we have created differs from the mechanised, post-Industrial Revolution era of homo faber and presents new conundrums. In this context, human action no longer fabricates objects but instead generates natural processes and channels them ‘into the human artifice and the realm of human affairs’.Footnote 19 It unleashes a possibly ‘endless new chain of happenings whose eventual outcome the actor is utterly incapable of knowing or controlling beforehand’.Footnote 20
In this article, AI encompasses a rapidly evolving family of technologies comprising machine-based systems that—for explicit or implicit objectives—infer from the input they receive ‘how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments’ with which they interact.Footnote 21 Machine learning models have made it possible for AI to simulate human intelligence by swiftly evaluating data and inputs, generating new data, and even possessing the ability to self-program and alter their own codes.Footnote 22 Natural language processing (NLP) is used to understand, extract, and use key information from text. Building on NLP, large language models (LLMs) are computing systems loosely inspired by the neurons in the brain, creating artificial neural networks (ANNs).Footnote 23 In the form of language models, most recently, AI has been able to generate human-like text based on the input it receives, enabling natural-sounding conversations and providing responses.Footnote 24
The clean energy sector is taking early steps to incorporate AI,Footnote 25 which has the potential to decarbonise as well as democratise energy systemsFootnote 26 while contributing to the UN-developed Sustainable Development Goals (SDGs).Footnote 27 On the other hand, AI’s dark sides are increasingly emerging as a result of its energy consumption and carbon footprint,Footnote 28 its built-in biases,Footnote 29 and its inability to truly understand meaning, operating more as a stochastic ‘parrot’.Footnote 30 Because LLMs can reorganise without any grounding and awareness of reality, they are not intelligent in the sense that they possess the capacity to understand. Rather, LLMs are endowed with a semblance of agencyFootnote 31 and process immense volumes of data, which remain only shadows of reality, much like in Plato’s allegory of the cave.Footnote 32
This article proposes a baseline of three principles for justice-centred AI, particularly in energy matters, from an Arendtian perspective. The turn to Arendt for conceptualising norms rests on two reasons. First, in Arendt the law is constitutive of a political community as it ‘creates first of all a space in which it is valid, and this space is the world in which we can move with freedom’.Footnote 33 Harking back to the etymology of the Greek term for law, nomos, Arendt calls it a space within which defined powers may be legitimately exercised.Footnote 34 The law should be constitutive—centring on constitutional and empowering rules, thereby opening new avenues for the plurality of human conduct—rather than solely regulative, hinging on the experience and content of prescriptions.Footnote 35 Second, Arendt’s perspective on humanity is ‘earthbound’ rather than human-centred, opening a more ecological set of principles to integrate AI technology within energy systems.Footnote 36
In this article, energy justice is the ‘goal of achieving equity in both the social and economic participation in the energy system, while also remediating social, economic, and health burdens on those historically harmed by the energy system’.Footnote 37 Ultimately, the goal is to trigger some further thinking, willing, and judging, in Arendt’s sense, around digital constitutionalism ‘as the embodiment of the limits to the exercise of powers in a networked society’.Footnote 38 Digital constitutionalism is poised to emancipate the debate on law and technology from the shackles of a technocratic, intellectual property, or solely privacy-bounded perspective,Footnote 39 particularly in the European Union. At the same time, no magical solution to harness AI for energy justice can be offered. More realistically, as for all types of justice, energy justice is a receding horizon,Footnote 40 slipping away as soon as it appears close. The quest for justice is not likely to receive an answer even in ‘just societies’, as societies are ‘just’ only insofar as they continue to question their level of justice.Footnote 41
While this article emphasises the growing power of corporate actors in AI-enabled energy systems, historical experience cautions that the concentration of power within the state has often posed equal, if not greater, risks to democratic governance. This dual concern reflects Arendt’s own diagnosis of totalitarianism, where the entwinement of bureaucratic rationality and technological systems enabled unprecedented forms of domination.Footnote 42 From a republican perspective, the relevant concern is not the identity of the actor, state, or corporate but the existence of arbitrary power capable of interfering with freedom.Footnote 43
Overall, this article is structurally limited by the realisation that AI governance is an emerging field of study. In fact, the regulation of AI unfolds under conditions of epistemic and normative uncertainty: the technologies under the AI umbrella keep evolving, knowledge of risks remains fragmented,Footnote 44 many harms are emergent and difficult to anticipate, and regulatory frameworks often institutionalise rather than resolve uncertainty.Footnote 45 At the same time, efforts to embed human values into AI systems may raise further challenges, as values are plural, context-dependent, and subject to contestation.Footnote 46 Accordingly, regulatory proposals should be understood not as definitive solutions but as iterative and revisable interventions within complex socio-technical systems.
III. The constitutive gap for justice-centred AI in energy matters
A. Introduction
A central challenge for justice-centred AI in energy contexts is what I term the ‘constitutive gap’. This gap reflects the distance between the democratic ideal of collective deliberation and the lived reality of individuals no longer ‘acting politically’ but ‘merely behaving’ as economic producers, consumers, and dwellers.Footnote 47
In the energy context, AI is a double-edged sword, poised to either perpetuate or possibly improve the current centralised and insufficiently democratic decision-making processes for energy systems.Footnote 48 Artificial neural networks and expert systems have been used for over 30 years in the energy sector, optimising the efficiency of several tasks and becoming the most utilised digital technology in electricity systems.Footnote 49 On the other hand, companies in the energy sectors, depending on their development and deployment of AI systems, are often deemed ‘safety-critical systems’, namely high-risk AI systems whose failure would put the health and well-being of citizens at risk.Footnote 50
Even the most comprehensive regulatory framework on AI, the EU AI Act, tackled the role of citizens through the means of the administrative state while missing the opportunity to fully recognise individuals’ rights.Footnote 51 On the one hand, the EU AI Act has set the highest bar in AI regulatory requirements through proactive governance,Footnote 52 both content and institution-wise. Content-wise, it has set forth a ‘risk-based’ approach, whereby the higher the risk of causing societal harm, the stricter the rules. Institution-wise, pursuant to the AI Act, an AI Office has been established within the EU Commission to ensure enforcement, as supported by an AI Board of member states’ representatives, a scientific panel of independent experts, and an advisory forum for stakeholders to provide technical expertise to the AI Board and the Commission, which adds to national enforcement authorities. On the other hand, the Achilles’ heel of the AI Act, beyond its level of generality regarding AI technologies and domains of application, is its effective enforcement, which is based on enforcement by public authorities alone.Footnote 53 Member states will decide on penalties, and the main guidance point from the AI Act is the imposition of a range of administrative fines.Footnote 54 Only high-risk AI systems are required to carry out a fundamental rights assessment, which may be judicially interpreted as conferring enforceable rights onto natural persons or groups.Footnote 55
The wide use of AI by energy actors makes the lack of basic statistics on the quantity and structure of electricity market agents even more striking.Footnote 56 Meanwhile, the literature is patchy regarding citizen participation in centralised systems and representational forms of democratic governance.Footnote 57 Such gaps raise questions on how to achieve more transparency across the actors tasked with adopting and enforcing AI systems in the energy space. To bridge the constitutive gap between current energy systems and the democratic ideal of collective deliberation, at least two re-conceptualisations are needed: the meaning of AI and its principled regulation.
B. The conceptual grounding of justice-centred AI
The vast array of qualifications of AI—trustworthy, human-centred, ethical, explainable, responsible, and fair—fail to include a justice component. Justice-centred AI is a relatively overlooked concept, whereby some scholars and leaders in AI ethics have focused on ensuring that AI systems promote equity, inclusion, transparency, accountability, the ethical use of data, representation, participation, the correction of historical biases, redress, and remedies.Footnote 58 A growing critique suggests that embedding ethical values into AI systems risks introducing distortions into their mathematical structure, potentially generating opacity and unpredictability.Footnote 59 However, this view presupposes a neutrality of computational systems that has been extensively challenged in the philosophy of science,Footnote 60 as well as in mathematicsFootnote 61 and law.Footnote 62 Rather than introducing values ex post, AI systems are already shaped by implicit normative choices of data, models, thresholds, and objectives. The new form of agency introduced by AIFootnote 63 is intrinsically shaped by goals and thus carries an inherent axiological dimension. Artificial agency, in fact, emerges from the interplay between programmed objectives and learned behaviours, constituting ‘a computational, goal-driven form of agency defined by human purposes’.Footnote 64 In this sense, AI agency marks a shift away from biologically grounded purposiveness towards engineered goal-directedness.Footnote 65 The question, therefore, is not whether values should be embedded but how they can be made explicit, contestable, and subject to democratic judgement. In the specific domain of energy law, the need for justice-centred AI becomes particularly acute. Energy systems are deeply intertwined with questions of access, affordability, sustainability, and governance, all of which are central to energy justice.
Following Thomas Franck, fairness is constituted of a procedural aspect—proper process—and a substantive aspect—distributive justice.Footnote 66 The procedural and substantive aspects are intrinsically related. In fact, legal systems are perceived as fair so long as the rules satisfy ‘the participants’ expectations of a justifiable distribution of costs and benefits’ and ‘the rules are made and applied in accordance with what the participants perceive as the right process’.Footnote 67 In particular, distributive justice favours change.Footnote 68 In the contested arena of values and their trade-off in a pluralistic society, distributive justice is also the foreground for deliberation, where ‘[h]eterogeinity and interpretative conflict’ are a resource, rather than a barrier, to deliberative problem-solving.Footnote 69
Distributive justice can be particularly challenging and difficult to incorporate into legal regimes, including integrating AI into energy matters. Initially, the environmental movement sidelined distributive justice as a distraction,Footnote 70 operating under the assumption that environmental policies would automatically yield a better environment for all. Since the 1980s, one of the most notable legal developments has been the rise of the environmental justice movement. At that time, less powerful communities realised that they were disproportionately impacted by environmental degradation but had limited possibilities to influence relevant decision-making processes. They thus resorted to legal action. Around the same time, the World Charter for Nature, adopted by the UN General Assembly in 1982, emphasised ecological justice by ensuring that all individuals have the right to seek legal redress for environmental damage or degradation. It acknowledged state sovereignty over natural resources while promoting public participation in environmental decision-making, thus allowing individuals to be recognised through actual involvement in and impact on decisions affecting their environment.Footnote 71 In the mid-2010s, the concept of climate justice was incorporated as a subset of ecological justice, addressing the disproportionate burden of climate change impacts on poor and marginalised communities.Footnote 72
Energy justice emerged in an academic context in 2010.Footnote 73 In 2015, one of the first comprehensive works on energy justice defined it as a mechanism poised to achieve procedural and distributive justice.Footnote 74 Rather than socio-technical fixes, energy justice requires transformative politics.Footnote 75 Instead of simply decarbonising energy systems,Footnote 76 energy justice should facilitate the participation of interested communities in the design, function, and ownership of energy systems,Footnote 77 particularly those devoid of the opportunities and the right to act.Footnote 78 Ultimately, like energy justice and other conceptions of justice, justice-centred AI in energy matters would need to encompass procedural, substantive, and recognition aspects, ensuring that algorithmic decision-making in the energy sector advances, rather than undermines, democratic ideals and equitable outcomes.
C. Constitutive principles: towards principled regulation
As noted earlier, constitutive principles, rather than haphazard regulation,Footnote 79 are highly needed to bridge the constitutive gap between current energy systems and the democratic ideal of collective deliberation. Pursuant to the modern constitutional tradition, such principles are constitutive in the sense of a constituent power reclaiming agency and self-government in a landscape shaped mainly by others.Footnote 80 Artificial intelligence makes what Nico Krisch calls the post-national sphere even more vivid, where ‘the structure of governance is the result of multiple interacting moves, organic growth and choices of powerful actors’.Footnote 81 It is challenging to determine the ‘who’ and the ‘what’ of political outcomes in the resulting networks of actors where the category of fate, not of constitution, seems to rule.Footnote 82 In the digital age, power has not simply shifted from states to corporations. It has been reconfigured through the rise of digital platforms and data-driven business models, which enable private actors to exercise forms of economic, epistemic, and infrastructural power traditionally associated with public authority.Footnote 83 Rather than being displaced, states have been entangled with the rise of corporate digital power, as they ‘created the conditions for extraction, legitimized its underlying logic, and often shielded it from democratic oversight’.Footnote 84 This development coexisted with a long-term expansion of state regulatory and administrative capacity since the post-World War II period. Ultimately, the commodification of human behaviour has given rise to surveillance capitalism, a new economic order reclaiming human experience as free raw material for its transformation for analysis and sales.Footnote 85 This model is distinct from, yet potentially mutually reinforcing with, government surveillance, and its expansion may contribute to more democratic disorder and de-institutionalisation.Footnote 86
If a power of new beginnings can nonetheless be rescued with regard to the ‘power couple’ of AI and energy (see Section I), it would be a pouvoir irritant—meaning, several critical stings irritating what can be deemed established constituted power.Footnote 87 From an Arendtian perspective, this type of constitutional power, albeit fragmented, allows for the ‘constitution of political freedom’, which creates new centres of power and new productive capacities among citizens.Footnote 88 Combining the ideas of Arendt with those of Alexander Hamilton, constitutive principles provide an opening—a ‘placeholder’—for societal aspirations to establish themselves out of reflection and choice.Footnote 89 The recent drive towards regulating the use of data and data-powered technologiesFootnote 90 is here described as one of the possible forms of digital constitutionalism, particularly ‘as a reaction to private norms and external interferences from other standards of protection’.Footnote 91
IV. Lifting the veil of the world
A. Regulating technology from an earthbound perspective
In Arendt’s terms, humankind has always been tempted to lift the veil of the future with the aid of technology, which can be broadly defined as the application of scientific knowledge for practical purposes.Footnote 92 Such will has accelerated with the idea of progress that emerged in the modern age and its drive to subject the world to its rule.Footnote 93 With the rise of science in the modern age, progress emerged as a new notion in the 17th century, becoming ‘the most cherished dogma of all men living in a scientifically oriented world’.Footnote 94 In the 18th and 19th centuries, with the transposition of progress from a concept of natural science to one concerning history and human affairs, exemplified by Hegel and Marx, a theory of ethics was ‘treated in the perspective of History and on the assumption that there is such a thing as Progress of the human race’.Footnote 95
Early on, Arendt underscored the first boomerang effects of science’s great triumphs, namely that the ‘truths’ of the modern scientific worldview can be demonstrated through our know-how—mathematical formulas and technological proofs—but cannot be fully understood in thought and speech.Footnote 96 Owing to this incapability of thought, we risk becoming enslaved not so much by our machines as by our know-how, becoming ‘thoughtless creatures at the mercy of every gadget which is technically possible’.Footnote 97 Presciently, Arendt underlined the difficulties of communicating the ‘truth’ of the modern scientific worldview, which is a sheer inability ‘to understand, that is, to think and speak about the things which nevertheless we are able to do’ through our action into nature.Footnote 98 In our technological world’s communicative gap, machines would overtake our brain, Arendt states, ‘so that from now on we would indeed need artificial machines to do our thinking and speaking’.Footnote 99
If technology cannot be fully thought of or spoken about, it strikes us in its apoliticality. The loneliness of human beings before technology emerges from the silence of traditional political and philosophical accounts of the world, which are themselves inadequate to interpret scientific and technological advances. Before the precipice of progress, we are also left without a banister towards meaningfulness. Insofar as human beings live and move and act in this world, they experience meaningfulness only because they can talk with and make sense of each other and themselves, which makes them political beings.Footnote 100
If the fear that human beings could be enslaved by technology has also characterised other authors,Footnote 101 Arendt’s view shifts the focus from humans to the Earth, proving particularly heuristic for the purposes of this article.Footnote 102 In Arendt, the Earth is the quintessence of the human condition, turning ‘earthly nature’ into a unique set of habitats where human beings can move and breathe without artifice. Through life thus intended, human beings remain earthbound, namely related to ‘all other living organisms’. Conversely, the human artifice separates us from this ‘mere animal environment’.Footnote 103 In this context, humankind seems to exchange existence—a gift given from nowhere, secularly speaking—for something that humankind has made itself, technology and the artifice more broadly.Footnote 104 Towards the same rebellious end of exchanging nature for artifice, this future human being has proved capable of destroying ‘all organic life on earth’, prompting questions on the following: which is the direction towards which we are using our new and scientific-technical knowledge? Far from being nostalgic for a pre-scientific era, Arendt tackled the ethics of technology because it is ‘a political question of the first order’, which cannot be left to professional scientists or politicians.Footnote 105
Accordingly, Arendt approached the technology question from an intrinsically ‘earthbound’ perspective. In this way, Arendt seems to foreground some of the tenets of the Anthropocene, whereby human and geological times intersectFootnote 106 in the ‘progress’ towards self-destruction, as mediated by technocratic dreams of world domination.Footnote 107 In response to the ensuing alienation from earthbound concerns, Arendt demands that science be reincorporated into the arenas of discussion and deliberation that underpin the political sphere.
B. AI regulation for the AI–energy power couple
Previously, I argued for the need to thoroughly regulate technology from an earthbound perspective, which counters the idea that technology develops independently and that law cannot steer innovation in technology.Footnote 108 This regulatory need is even more overt in AI regulation because AI is inherently difficult to regulate owing to its innovation ratio and the character of general-purpose technology, opening the floodgates to use and misuse. Artificial intelligence regulation breaks down into at least three macro areas concerning (1) spending (eg private investments and public funding); (2) competition, innovation, and ownership (eg intellectual property and reduction of rent extraction); and (3) ethics and safety (eg human rights, consumer protection, bias and discrimination, transparency, self-replicability, and explainability). Such macro areas are not orthogonal and create trade-offs, for instance, between private investments and fair competition or consumer protection and innovation. Because of insufficient regulation, AI as a new power source has yet to be socially or politically legitimised through effective governance.Footnote 109
The first national AI strategies appeared in 2017 and now number over 50, comprising over 930 policy initiatives across 71 jurisdictions.Footnote 110 Several national strategies align with the 2019 Organisation for Economic Co-operation and Development (OECD) AI Principles, updated in May 2024Footnote 111—the first intergovernmental standard on AI. The OECD AI Principles identify five complementary value-based principles for the responsible stewardship of trustworthy AI: inclusive growth, sustainable development, and well-being; human-centred values and fairness; transparency and explainability; robustness, security, and safety; and accountability. The OECD AI Principles recognise that AI has the potential to contribute positively to sustainable global economic activity while potentially having disparate effects within and among societies.Footnote 112 Accordingly, they emphasise the need for a well-informed, whole-of-society public debate to capture the beneficial potential of the technology while limiting the risks associated with it.Footnote 113 While the principles stop short of considering AI in energy systems and seem to conflate energy with climate and other AI sustainability issues, they importantly emphasise the need to empower stakeholder engagement.Footnote 114
The Council of Europe’s (COE) Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law was agreed upon by 57 states in May 2024 and is the first binding international treaty on AI. From the start, the aspiration has been for the convention to be adopted worldwide by non-COE states and to remain future-proof by not regulating technology,Footnote 115 which is a remarkable, laissez-faire policy statement. Notwithstanding, the convention reaffirms a commitment to a series of foundational human rights documents: the 1948 Universal Declaration of Human Rights, the 1950 Convention for the Protection of Human Rights and Fundamental Freedoms, the 1966 International Covenant on Civil and Political Rights, the 1966 International Covenant on Economic, Social and Cultural Rights, and the 1961 European Social Charter, as revised in 1996, as well as their respective protocols; the 1981 Convention for the Protection of Individuals with Regard to Automatic Processing of Personal Data and its protocols; the 1989 United Nations Convention on the Rights of the Child; and the 2006 United Nations Convention on the Rights of Persons with Disabilities.Footnote 116
Importantly, pursuant to Article 7 of COE’s the framework convention, each state party to the convention shall ‘adopt or maintain measures to respect human dignity and individual autonomy in relation to activities within the lifecycle of artificial intelligence systems’. As the other side of dignity, pursuant to Article 9 of the convention, each state party shall also adopt or maintain measures to ensure accountability and responsibility for adverse impacts on human rights. To fulfil the protection of dignity and human rights through accountability, state parties shall also impose accessible and effective remedies and procedural safeguards.Footnote 117
Another regional instrument, the EU AI Act, was the first comprehensive regional law on AI. It is meant to protect human rights and goes one step further by underscoring the need to reduce the consumption of energy and other resources by AI systems.Footnote 118 It also enshrines the possibility to create AI regulatory sandboxes that the European Commission would facilitate through delegated acts to avoid fragmentation across the Union.Footnote 119 Such regulatory sandboxes would enable a liberal, albeit controlled, use of personal data to develop certain AI systems in the public interest, such as for ‘energy sustainability’.Footnote 120 Notably, the AI Act acknowledges the specificities of the latter compared to environmental, climate, or biodiversity matters.Footnote 121
Beyond such regional instruments, national attempts have mushroomed without attaining a sufficient level of specificity regarding the power couple of AI and energy. What stands out is regulatory divergence, in particular, among the EU’s preventative approach,Footnote 122 the US digital laissez-faire,Footnote 123 and China’s ‘strong government’ approach whereby, in case of conflict, national interests are meant to prevail over individual citizen rights.Footnote 124 Ultimately, the risk of regulatory arbitrage is real.
At the global level, G7 or country-initiated summits have yet to tackle the risks and opportunities of AI in energy systems,Footnote 125 while UN efforts to coordinate AI policy seem more promising. After a Mexican-led UN initiative received little interest in 2017,Footnote 126 in March 2024, the first UN resolution on AI, proposed by the United States, was adopted by the General Assembly in a consensus vote. In particular, the resolution calls upon member states and other stakeholders to refrain from or to cease using AI systems that are impossible to operate in compliance with international human rights law or that pose undue risks to the enjoyment of human rights, especially of those who are in vulnerable situations. At the same time, the resolution draws an equivalence test between online and offline rights protection throughout the life cycle of AI systems.Footnote 127
Although the March resolution stops short of considering AI in energy systems, it reaffirms a commitment not only to the UN Charter and the Universal Declaration of Human Rights but also to Agenda 2030, thus underscoring the role of AI in enabling the SDGs.Footnote 128 The resolution thus entrusts international law with providing the appropriate basis for devising AI governance systems that are ‘interoperable, agile, adaptable, inclusive, responsive to the different needs and capacities of developed and developing countries alike and for the benefit of all’.Footnote 129
Last, it handed the baton of further international AI lawmaking to the Summit of the Future, where the Global Digital Compact was annexed to the Pact for the Future in September 2024.Footnote 130 In the Digital Pact, references to AI sustainability are few and far between, mainly revolving around Objective 1(e), which addresses sustainability across the life cycle of digital technologies with explicit reference also to SDG 7 on clean, affordable energy and SDG 13 on climate.Footnote 131
Overall, the regulatory frameworks discussed here provide some specificity regarding the sustainability requirements of AI systems. However, likely owing to gaps in international law on AI in energy systems, they tend to overlook the implications of AI from an energy justice perspective, as well as the role of communities and their deliberations in AI-enabled energy systems. In this respect, the reviewed frameworks seem to epitomize a trend in international law where transparency has increasingly supplanted accountability, participation, and even distributive justice.Footnote 132
V. Three justice-centred principles for AI in energy systems
A. Introduction
I have argued that a justice-centred approach to AI is necessary, as technology evolves far more rapidly than regulation, while regulatory purposes remain so broad that they risk diverting efforts towards overly abstract initiatives. Existing legal and policy frameworks therefore provide little traction for addressing the concrete practices through which AI shapes society. What is missing is not more regulation at a higher level of generality but a principled orientation capable of guiding both governance and evaluation. In short, we lack a normative framework of justice with which to evaluate the practices that constitute AI.Footnote 133
The following three constitutive principles can offer a starting point for prospective discussions on the communitarian and individual role of AI in energy systems. In particular, these constitutive principles aim to illuminate how regulatory frameworks can harness technology for localised forms of political activity to ultimately enhance energy justice in AI contexts. After introducing the first principle, by which AI must be treated strictly as a means, never as an autonomous end, and the second principle, hinging on fiduciary duties to strengthen corporate accountability, the third principle approaches the role of AI in energy systems through the right to have rights, which would be relevant in at least three ways: as an ‘irritant’ to centralised powers at both public and private levels; as a reinforcer of human dignity against the over-exploitation of nature; and as an enabler of democratic deliberation in the increasing opposition between culture and technology.
B. First principle: the categorical imperative reversed
The first principle of justice-centred AI consists of a reversed categorical imperative. Informed by Kant’s second formulation of the categorical imperative, or principle of humanity, we ought to ‘treat humanity, whether in (our) own person or in that of any other … as an end withal, never as means only’.Footnote 134 Pursuant to this principle, providers and deployers of AI should not only treat humanity as an end but also treat AI as a means rather than an end—an approach that is not always observed. The relevance of the categorical imperative in AI is owing to its foundational role in the emergence of human rights in the 20th century and its permeation of the corpus iuris of international law into the present day,Footnote 135 as well as to the existence of strikingly similar principles across cultural traditions and in secular humanism.Footnote 136 Further, in its Renaissance tradition, dignity is the power to opt for different paths of development—”thou art confined by no bounds”—rewarding those who develop their intellect.Footnote 137 At the same time, human dignity entails the capacity inherent in human nature to assume obligations vis-à-vis others, including nature.Footnote 138 Importantly, human dignity and individual autonomy are also enshrined in the COE’s AI Convention.Footnote 139 Overall, the practical significance of the categorical imperative in AI matters is self-knowledge—knowledge of ourselves as rationally efficacious and responsible agents.Footnote 140
Conversely, a distinct narrative seems presently at work in AI matters. According to this, non-human systems are analogous to human minds, and with sufficient training human-like intelligence can flourish,Footnote 141 meaning that our humane capabilities are outmoded and non-essential for superhuman intelligence.Footnote 142 According to this narrative, in the inexorable course of progress, AI is a superhuman power beyond our control and, thus, ungovernable, which greatly exaggerates its risks and distracts from other existential threats, notably climate change and the energy transition.Footnote 143 The implication is regulatory laissez-faire insofar as new AI inventions, ideally subsidised by public investment, can ensure safety from the AI bogeyman threatening human eradication.Footnote 144
Differently, from an Arendtian perspective, people’s dignity demands that they are seen, every single one in their particularity, which contradicts the very idea of progress as the law of the human species.Footnote 145 Along these lines, the principle of humanity strengthens self-governance—namely the capability to think, will, and judge for oneself how best to live—which underpins the civil and political liberties guaranteed by constitutions, international law, and democratic life more generally.Footnote 146 As Shannon Vallor expounds, such faculties of self-determination allow for decisions about how best to live to be ‘made in political cooperation with those whose fates are intermingled with ours’Footnote 147—not by AI. Technology can respond to the how but not to the why and the for what: why we make one choice and not another.
In the ‘power couple’ of AI and energy, one of the practical consequences of the reversed categorical principle is a set of constitutive counter-powers to citizens. Notably, when AI is deployed in energy systems, individuals should be the masters of the data that are extracted from them. This first categorical principle can, for instance, entail introducing contract clauses on data sovereignty and security design and ensuring data encryption to preserve citizens’ privacy.Footnote 148 This counter-power also includes applications to community energy, a global phenomenon by which energy initiatives are owned, developed, decided upon, and often managed by local communities rather than corporate actors.Footnote 149 Presently, however, even some of the most ambitious regulatory frameworks on energy communities at the EU level fail to ensure an effective governance model warranting local ownership and deliberation.Footnote 150
The reform proposed would be a rights-based model of informational self-determination, grounded in fundamental rights, rather than in property rights, and enforced through regulatory obligations. This approach is clearly reflected in the landmark Google Spain v AEPD, where the Court of Justice affirmed that control over personal data derives from fundamental rights rather than proprietary claims.Footnote 151 Differently from the EU AI Act, the GDPR already establishes a comprehensive set of data subject rights, including access (Art 15), erasure (Art 17), and portability (Art 20), which together operationalise a form of individual control over personal data.Footnote 152 The Data Act extends this logic beyond personal data by introducing rights of access and sharing with respect to data generated by connected devices, including in the energy sector, thereby addressing asymmetries between users and data holders in increasingly data-driven infrastructures.Footnote 153 Broadly stated, in EU law, digital rights amount to the opportunity to be seen and heard in the public realm, in an Arendtian sense,Footnote 154 or to withdraw from it, expounding one’s positive and negative freedom as digital sovereignty.
Although the concept of European digital sovereignty was foreshadowed in the Digital Services Act and furthered in the Digital Markets Act and the Data Governance Act,Footnote 155 its embodiment appears legally unfinished and unsatisfactory in the EU AI Act, which reflects a shift from a value-based debate to a risk-centred and mainly safety-oriented regulation,Footnote 156 resulting in a framework that lacks the normative depth required to guide long-term technological development. In this sense, the proposed principle seeks to complement the existing legal framework by addressing a residual gap: resolving the collective and systemic dimensions of data governance, particularly in contexts such as AI-driven energy systems where value is generated through aggregation, inference, and cross-sectoral data flows.
The need for stricter governance rules, notably on community energy, would be an ‘irritant’ to what presently appears to be the corporate capture of energy communities whenever private investors (eg investment funds) secure majority equity and capture the public incentives supporting such initiatives without any guarantee that crucial AI-harvested data, decision-making power, and sufficient ownership remain under the control of the actual prosumers, namely those who produce, store, or consume energy.Footnote 157
Approaching the role of AI in energy systems through a reversed categorical imperative would be relevant in at least three ways. First, the ‘circle of dignity’ underlying fundamental rights aligns with the foundation of contemporary international law, and, as argued by Ginevra Le Moli, Kantian dignity relates to respect and is relational rather than self-referential.Footnote 158 On this point, the reversed categorical imperative also aligns with energy justice metrics, in particular with respect to participation and to better sharing of economic and social benefits with communities that previous energy systems have neglected, thanks to technology, instead of despite it.Footnote 159 Second, stricter governance rules for AI in energy systems are also required to avoid the captive phenomenon of socialising bailouts and privatising gains,Footnote 160 which has long been a feature of late capitalism and appears to be ramping up in AI financing.Footnote 161 By allowing corporations to profit—accumulating public money without strict guarantees—from AI-extracted data, any regulation would treat AI and its experimentations as an end rather than a means. Third, from an Arendtian perspective, ‘human dignity needs a new guarantee which can be found only in a new political principle, in a new law on earth, whose validity this time must comprehend the whole of humanity while its power must remain strictly limited’.Footnote 162 To take this insight seriously means recognising that dignity cannot rest on abstract declarations alone. Rather, it requires embodiment in a framework of rules that are not only strict (see Section V.C) but also enforceable (see Section V.D).
C. Second principle: diligentia diligentis, or the diligence of a reasonably prudent person
The second principle of a justice-centred AI consists of a clear set of fiduciary duties to be imposed on AI providers and deployers. Fiduciary duties generally provide a legal infrastructure for relationships spanning the care, custody, and administration of persons, property, and organisations.Footnote 163 Fiduciary duties arise where one party is entrusted with discretionary authority over another’s interests and is legally bound to exercise that power solely for the beneficiary’s benefit.Footnote 164 These duties appear across legal domains, from classic private law relationships—such as trustee–beneficiary or agent–principal—to corporate and public governance contexts.Footnote 165
The first conceptualisation dates to Roman law: this duty, diligentia diligentis patris familiae (the diligence of the diligent good father), in the Corpus Juris Civilis later evolved into canon law and Medieval English case law.Footnote 166 Although fiduciary duties are widely debated,Footnote 167 at least three can be identified: the duties of loyalty, prudence, and impartiality.
The duty of loyalty requires acting with undivided fidelity, untainted by self-interest. For AI providers and deployers, duties of loyalty would appear, such as the obligation to devise safety systems of precautionary action and disclosing in plain language the potential and risks of using AI-extracted data in energy systems. In this respect, regulators are starting to govern information markets to ensure the sufficient quality and volume of information.Footnote 168 Such regulation can counterbalance structural asymmetries owing to an unregulated information market, avoiding distortions.Footnote 169 As a result, the transparency layered through disclosure duties may initially substitute for public participation,Footnote 170 especially when the general public has few means to enforce entities’ disclosure duties.
The duty of prudence requires fiduciaries to act with caution, care, and diligence in managing the systems for their beneficiaries.Footnote 171 In particular, for AI providers and deployers, duties of care under the duty of prudence would include, for example, the obligation to examine the factual basis and ecological impact of AI applications in energy systems, including the carbon budget associated with the energy intensity managed and optimised through such systems. This rule would also preclude providers and deployers from merely replicating market investments and applications that are ultimately financed by energy consumers, thereby privileging short-term profits while mispricing sustainability risks.Footnote 172
The duty of impartiality refers to good-faith efforts to identify, respect, and balance diverse interests among beneficiaries when carrying out fiduciary responsibilities.Footnote 173 In particular, for AI providers and deployers, duties of impartiality would flesh out, for instance, in the obligation to consider not only the stakes of present generations but also the intergenerational duty that, in principle, should underpin impartiality.Footnote 174 In this sense, impartiality entails ‘taking the viewpoints of others into account’.Footnote 175 Equity would provide ‘standards for allocating and sharing resources and for distributing the burdens of caring for the resources and the environment in which they are found’,Footnote 176 including energy resources, with the principle of equity between generations underlying the very notion of sustainable development.Footnote 177 A notable example would be the requirement that AI providers and deployers work on a more ecological principle of reducing LLMs’ carbon footprint while maintaining the same level of effectiveness.
Approaching the role of AI in energy systems through fiduciary principles would be relevant in at least three ways. First, it incorporates several legal traditions and cultures. In fact, comparable fiduciary duties can be found in both common and civil law traditions, usually complemented by duties arising from other legal sources, such as statutes or contracts.Footnote 178 At the same time, fiduciary principles would offer a more concrete policy response to the present call to make AI providers and deployers accountable. A notable framework is the OECD AI Principles, by which all AI actors should apply a systematic risk management approach to each phase of the AI system life cycle, including across their value chain, based on their roles, context, and ability to act.Footnote 179
Second, fiduciary duties, as described earlier, require a type of due diligence in connection to sustainability that has long existed in environmental and energy matters but has been consistently overlooked in AI regulatory attempts. Sustainability due diligence should cover limits on the use of planetary resources, social issues—notably human rights protection—and economic and governance issues, with recurring assessments at an interval of at least three years—potentially more in case of business changes—across the global value chain.Footnote 180
Third, fiduciary duties would offer concrete procedural steps to fulfil the nexus of human rights with environmental principles in energy matters,Footnote 181 notably prevention, sustainability as an umbrella principle, and the intra- and intergenerational equity principles, which feature among the core principles of energy justice in international law. From an Arendtian perspective, the role of procedures as irritants that countervail the power of corporate entities and bureaucratic societies can humanise the ability to handle things that are not yet properly encompassed by the faculty of the will.Footnote 182 Within the myth of progress, when driven solely by profit, businesses have revealed ‘the utter amorality of the profit motive and its indifference to consequences’, as Beth Stephens recalled more than 20 years ago in reference to companies that sold revolutionary data management systems to the Nazis.Footnote 183
Overall, the power couple of AI and energy can transform corporate accountability from current conversations on transparency to more concrete, outright fiduciary duties arising from human rights and environmental principles. Nonetheless, fiduciary duties should be painstakingly designed as enforceable. An Arendtian perspective would deny the factual validity of human rights corresponding to fiduciary duties not per se but to the extent that human rights do not entail any enforceable praxis on their own.Footnote 184 In this sense, human rights are, in principle, inalienable if a prerequisite is secured—namely the right to possess rights and to enforce them as a citizen of a political community—as we will see in the following description of the third principle for justice-centred AI (Section V.D).
D. Third principle: the right to have rights
The third principle of justice-centred AI is comprehensive. It is the right to have rights, an Arendtian syntagm meant to secure human dignity that consists of the right to enforce one’s rights while engaging in meaningful political action.Footnote 185 Arendt’s argumentation for this right springs from the historical experiences of the 20th century, when human rights, ‘supposedly inalienable, proved to be unenforceable even in countries whose constitutions were based upon them—whenever people appeared who were no longer citizens of any sovereign state’.Footnote 186 Although Arendt directed her main critique to the calamity of the rightless and stateless, her proposal of the right to have rights remains striking in its normative content. According to Arendt, political action in a situated community is the means to exercise freedom, while the right to have rights is denied whenever individuals no longer belong to a community.Footnote 187
Along these lines, the right to have rights springs from participation as members of a group whereby we ‘guarantee ourselves mutually equal rights’.Footnote 188 Consequently, community is never a given. It is a construct that citizens should actively design and maintain through opinions, contracts, and promises.Footnote 189 In Arendt, the civic bond is provided by neither common interests nor the common good. Instead, it is the common world where we can disagree and dissent—that is, debate about what is between us, the inter-esse.Footnote 190
In this world outside and between us, public interest is the measure for public life, securing the permanence that lies in the public sphere and materialises in institutions.Footnote 191 If any social contract can be envisaged in Arendt, it would be a horizontal contract lifting each individual out of isolation while also limiting their individual power in view of relational power, or action-in-concert, in a shared world.Footnote 192 This essentially horizontal and political understanding of the separation of powers led Arendt to understand the separation of power as implying more power—in the sense of more loci of power—to buttress a robust political community.Footnote 193 The nation-state’s notion of sovereignty, a dangerous megalomania originating from absolutism,Footnote 194 cannot ensure the separation of powers as a counterbalance of powers through the multiplication of its loci. In Arendt, ‘there can only be democracy … where the centralisation of power in the nation-state has been broken, and replaced with a diffusion of power into the many power centres of a federal system’ of sorts.Footnote 195
In Arendt, human freedom is a matter not of metaphysics but of fact.Footnote 196 It is experienced primarily not in will and thought but in action, requiring a political sphere for such action-in-concert.Footnote 197 To operationalise freedom as an essentially political phenomenon, Arendt turns to two forms of direct political participation: Thomas Jefferson’s ‘ward system’, as inspired by the town hall meetings of the American Revolution, and the citizens’ councils mushrooming in 19th and 20th-century Europe.Footnote 198 Rather than the demise of modern representative democracies, Arendt emphasises the need to embody effective democratic practices through localised forms of political engagement. The latter, as ‘little republics’, would institutionally cement the pouvoir constituant of the people and prevent centralised powers from acting on any tyrannical tendencies.Footnote 199 The right to have rights brings the law to the centre by requiring spaces that preserve the pre-political condition of the existence of a common world of inter-esse.Footnote 200 The need for ‘little republics’ also shows that our society often aims to leverage AI to increase our control over an uncertain future, while AI can—paradoxically—reduce our agency over the future, thus exposing the myth of progress on which modernity has so largely relied.Footnote 201
Transposing Arendt’s reflections on the ‘power couple’ of AI and energy, energy consumers face the same paradox of stateless individuals whenever the nation-state does not guarantee their right to have rights. In the alienating world of energy consumers, the right to have rights would require not only an opportunity to impact the lawmaking process but also a ward system whereby, through speech, individuals can make choices about their freedom. In this ward system, humans would also appropriate and, as it were, dis-alienate ‘the world into which, after all, each of us is born as a newcomer and a stranger’.Footnote 202 Decentralised forms of decision-making and self-government are required in energy matters, all the more so when AI systems are engaged, thus demanding the painstaking work, yet to happen, of institutional translation from democracy in paper to democracy in practice. Such work would reboot political freedom not only within energy communities but also through less formalised ways of debating AI and energy matters.
In the power couple of AI and energy, one of the practical consequences of the right to have rights is the possibility for AI to accelerate discussions and the implementation of ecological citizenship beyond the value of transparency alone. By supplanting the universalism customarily associated with citizenship, ecological citizenship can be understood as providing spaces of ‘insurgent’ citizenship.Footnote 203 In this way, laws would offer practical pathways to thematically cope with the unfulfilled promises of representative democracy through the creation of new, specialised spaces of deliberation, such as energy communities where the institutional translation of the right of rights ought to be taken more seriously.
Truly, meaningful political participation cannot be mandated by law. Overly prescriptive forms of participation risk formalism, reinforcing passivity and encouraging citizens to externalise responsibility to the state. The function of law, therefore, is not to compel participation but to enable and structure it, by lowering barriers to engagement, creating accessible fora, and ensuring that participation has tangible effects on decision-making. In this context, approaching the role of AI in energy systems through the right to have rights would be relevant in at least three ways. First, the right to have rights would be an ‘irritant’ to centralised powers at both public and private levels, which constitute the large majority of energy systems and which, in Arendt, always risk stagnating in the absence of any counterpower.Footnote 204 Second, it aligns with a possible constitutive stage of human dignity, as foreshadowed by Ginevra Le Moli. In fact, it limns a more ecocentric understanding of the knots between humans and nature against the over-exploitation of nature.Footnote 205 In this further circle of dignity, dignity rises to a more political form where ‘[p]lurality is the law of the earth’Footnote 206 because ‘[m]an is not merely conditioned by his environment; he conditions the environment, and the environment then conditions him in turn’.Footnote 207
Third, among the suggested three principles, the right to have rights opens up the conditions for the more legitimate design and use of AI in energy systems because it is the most political among the three principles. In fact, it rests on the assumption that the direction of technology should be the object of democratic deliberation. In the increasing opposition between culture and technology,Footnote 208 a common world of deliberation, promises, and contracts can shape a type of culture that better conciliates technology with nature.
In the words of Romano Guardini, under whom Arendt studied in her university education, this premise of deliberative spaces is not anti-technological, but it would ensure that ‘the process of technology worldwide will really achieve the great things that it can and should’.Footnote 209 That technology brings a new measure of freedom is, in principle, a gain; the ‘value of freedom, however, is not fixed solely by the question “Freedom from what?” but decisively by the further question “Freedom for what?”’. In fact, in absolute modernity, conflicts unfold between ‘the embodiment of ideas in social practice (ethos) and the rationale behind practice (logos)’.Footnote 210
VI. Institutionalising justice-centred AI principles
A. Introduction
Previously, I explored the constitutive gap between the democratic ideal of collective deliberation and the reality of individuals behaving as energy consumers. To help bridge this gap, I further argued that international law could be a gap-filler by offering a baseline of three principles for justice-centred AI in energy matters: the categorical principle reversed, a set of fiduciary principles through the diligentia diligentis (the diligence of a reasonably prudent person) standard, and the provision of the right to have rights. Finding these principles in need of institutionalisation, I here propose three methods to embed them through both centralised and decentralised means. What I mean by this is that institutionalisation initiatives are needed across levels of government, spanning local and global governance fora, the latter embodied in the OECD AI Policy Observatory and the UN, given the newly inaugurated season of AI resolutions and the 2024 Pact for the Future.
B. Institutionalising the reversed categorical imperative
To review, the first principle on the categorical imperative reversed holds that providers and deployers ought to treat humanity as an end and AI as a means. To uphold this humanity principle and its underlying tenet of human dignity, one can envisage at least three institutional arrangements. First, knowledge hubs would foster AI literacy, often recalled in policy documents, albeit at a high level of generality and without factualisation.Footnote 211 Because AI affects society at large and is rapidly evolving in unpredictable ways, AI literacy hubs would nurture reflection and nuanced scepticism in such a way that the emergence of machine intelligence is steered through critical intelligence towards planetary sapience in the service of a viable long-term future. At the same time, AI literacy hubs can counteract the frightening feeling of ‘solipsistic freedom’—‘the “feeling” that my standing apart, isolated to everyone else, is due to free will, that nothing and nobody can be held responsible for it but myself’.Footnote 212
Artificial intelligence knowledge hubs have already been proposed, albeit mainly in the Nordic region in Europe,Footnote 213 and would align with the United Nations Educational, Scientific and Cultural Organization’s (UNESCO) guidelines and frameworks for ethical AI.Footnote 214 Artificial intelligence literacy hubs would also help reconceptualise the public sphere for the ‘electro-iconographic societies’ of late capitalismFootnote 215 and help the youth confront the iconographic medium of the electronic means of representation,Footnote 216 including through online activities.
Second, AI companies would be required to embed safety institutions in their organisational structure. Safety institutions are presently rare and under-incentivised, or used to ‘seek government and philanthropic funding to invest in technical study of the unknown, longer-term risks of future “frontier” models that could be more dangerous than those we have today’.Footnote 217
Three, AI bias should be the object of stricter discussions and regulations concerning training materials and output control. Artificial intelligence bias counters the very principle of dignity by lumping individuals under discriminatory labels, practices, assumptions, and assertions, thus worsening the algorithmic echo chambers, which thoroughly diminishes our political sense of a shared reality and enables new forms of political manipulation.Footnote 218 The COE’s AI Convention also encourages equality and non-discrimination protections.Footnote 219
Overall, from such institutionalisations of the principle, a more solid understanding would follow about how AI systems function and how they can be used, which would in turn increase the acceptance of AI.Footnote 220 Understanding is needed not only of the known risks and uncertainties posed by AI but also of the unknowns attached to the technology, even for AI developers and researchers.Footnote 221 Further, these reform suggestions would widen the chances for humans to engage in reflection in a non-cognitive, non-specialised sense as a natural need of human life, thus becoming a prerogative of not only a few so-called scholars and experts but all humans, regardless of cultural, educational, and social conditions.Footnote 222
C. Institutionalising diligentia diligentis
To review, the second principle rests on the diligence required of AI providers and deployers through fiduciary duties. It has the potential to tackle the risks associated with the use of AI in energy systems, notably ‘lack of transparency, decline of human autonomy, cybersecurity, market dominance, and price manipulation on the electricity market’.Footnote 223
To uphold fiduciary duties and the underlying tenet of diligentia diligentis, one can envisage at least three institutional arrangements. First, fiduciary duties should be discussed and made mandatory in order to limit managerial autonomy for the self-preservation of energy system functions.Footnote 224 This approach will replace non-mandatory and individually defined standards of care with measurable metrics for which AI providers and deployers are accountable,Footnote 225 with an enhanced role for AI safety institutes and third-party controls. Accordingly, fiduciary duties can become enforceable functional duties of disclosure, auditing, and control processes.Footnote 226
Second, beyond this governance level, a set of due diligence processes should be detailed to further tackle the growing horizontal relationships connecting individuals with private digital companies capable of competing with public authorities.Footnote 227 Artificial intelligence due diligence would require AI developers to conduct mandatory pre-deployment sustainability assessments, including the impact of AI on both ecological and social values, notably fundamental rights.Footnote 228 For high-risk AI systems, such as energy matters, mandatory algorithmic impact assessments would be coupled with independent third-party audits throughout their life cycle. Algorithmic impacts would also be needed in order to ensure responsible deployment decisions and a safety profile across the AI value chain over time.Footnote 229 Given the scope for discretion in risk assessments and relatively limited experience in AI matters, when compared to privacy,Footnote 230 the EU AI Act requires guidance and disambiguation in this respect.Footnote 231
Third, and connected to the second point, the other side of the principle diligentia diligentis is enforceability, namely the institution of ‘stronger mechanisms for contestability, liability, and redress for avoidable and significant AI harms, to reinternalise the costs of preventable harm and developer negligence currently being imposed upon vulnerable publics’.Footnote 232 Current victims of AI have few paths of remedy and redress at their disposal, while such mechanisms have been key in building and sustaining the innovation and safety cultures of other industries, notably civil aviation, civil engineering, and pharmaceuticals.Footnote 233
Overall, the proposed systems of legal accountability, contestability, and redress would better incentivise AI developers and deployers to meet a high standard of careFootnote 234 in the face of current political failures to offer proper compliance incentives or effective, flexible, and iterative regulation.Footnote 235 Moreover, this approach would counter the mushrooming yet slight effectiveness of voluntary codes of conduct, which are upheld even by what has been deemed the stricter AI regulatory framework, namely the EU AI Act.Footnote 236 Further, stopping short of institutionalising limits to AI design and deployment, concerns would only increase with regard to the fate of the rule of law in an algorithmic society.Footnote 237
D. Institutionalising the right to have rights
To review, the third principle posits that individuals ought to be secured the right to enforce their rights while engaging in meaningful political action. Rather than warranting participation, the right to have rights guarantees the conditions for effective democratic practices through localised forms of political engagement. To uphold the right to have rights and the underlying fulfilment of human dignity in its more political form, one can envisage at least three institutional arrangements. First, public and private actors ought to recognise that the ‘decentred’ public sphere, including electronic media, is where most value debates are waged and ‘subaltern’ politics is carried out by previously excluded and marginalised groups.Footnote 238 As for more traditional forms of the public sphere, the task there would be ‘to challenge the secret logic of power, hierarchy, and domination’ in narratives and means.Footnote 239 At the same time, offline and online public spheres should be taken seriously and moderated to foster an ethics of the limit while new technologies induce a feeling of unlimitedness.
Second, it should be acknowledged that the demand for the right to have rights in the public sphere is particularly challenging because the public sphere is a nostalgic trope.Footnote 240 Democracy rests on the idea of an autonomous public sphere for self-governance through collective deliberation. There exists, however, a hiatus between this regulative ideal and ‘the increasingly desubstantialized carriers of the anonymous public conversation of mass societies’.Footnote 241 This hiatus turns the ideal of democracy into a constitutive fiction, causing constant anxiety.Footnote 242
Third, because the power of encoding thoughts into language is still the hallmark of our ‘unique cognitive liberty and the heart of our political capacity’,Footnote 243 the institutionalisation of the right to have rights ought to emphasise thought and speech as ‘the greatest interest and most distinctive achievement of man’.Footnote 244 As notably encouraged at the OECD, EU, and Nordic levels,Footnote 245 regulators can support safe ‘sandbox’ testing environments for thought and speech with regard to new AI solutions. Instead of including only expert knowledge (ie, researcher and innovator participation),Footnote 246 AI regulatory sandboxes should be informed by a ward system of multi-stakeholder debate and cooperative meetings built in federal units under a municipality, for example town council meetings built into federal units. A type of council meeting would also happen online and target the youth in order to inspire their self-empowerment and self-governance, for instance with culture—which alone is able to arrest and move us.Footnote 247
In Arendt’s constitutional scheme, such councils would play a deliberative and decision-making function ‘in opposition to the party system as an antidote to their electoral manipulation and depoliticising effect’.Footnote 248 Councils can discuss and attempt to solve problems through ‘incompletely theorised agreements’, which Cass Sunstein foreshadows as key to legal and political reasoning for their social stability function. They facilitate convergence, reduce the costs of disagreement, and help demonstrate humanity and mutual respect in a liberal democracy.Footnote 249 Agreements are incompletely theorised because participants agree on the results but not on the underpinning rationale.Footnote 250 Incompletely theorised agreements that reduce the cost of perpetual disagreementFootnote 251 are the best option for people with limited time and capacities,Footnote 252 making them widespread. These agreements serve as the foundations for both rules and analogiesFootnote 253 and are well suited to a world containing social dissensus on large-scale issues. At the same time, such agreements remain provisional and contestable, requiring iterative testing and refinement, which is an approach that aligns with experimental regulatory tools such as AI sandboxes in the EU AI Act.
Nonetheless, AI citizens’ councils are no guarantee of justiceFootnote 254 and would require a range of cases in which the principles are tested against others and refined,Footnote 255 which would align with the requisite for AI regulatory sandboxes. This type of agreement can also align with what has been called ‘algor-ethics’, namely the moderation of algorithms and AI programs pursuant to discrete values. Notwithstanding value pluralism, where ‘a single set of global values’ seems impossible to find, we can still retrieve shared principles governing the role of AI in our common world,Footnote 256 and incompletely theorised agreements can be a tool to such an end. While citizens’ councils may rely on digital infrastructures to enable coordination, their legitimacy cannot rest on purely virtual interaction. In Arendtian terms, meaningful political action presupposes a shared space of appearance rooted in the world. Accordingly, councils should be conceived of as ‘onlife’ institutions: digitally extended, yet anchored in situated deliberation, where participants engage with the material consequences of AI systems within their local environments. Citizens’ councils would function as standing deliberative bodies that scrutinise and inform in real time the design and deployment of AI technologies, especially of high impact, in line with the emerging field of participatory AI.Footnote 257
Within this architecture, councils would also operate as distributed sensing nodes, capturing localised environmental externalities and societal impacts of AI systems, thereby contributing to a polycentric model of governance grounded in context-sensitive knowledge.Footnote 258 In fact, AI governance in energy systems can be fruitfully understood through the lens of common-pool resource (CPR) dilemmas, albeit in a non-classical sense. While AI as such does not constitute a CPR, its underlying infrastructures and impacts, particularly compute-intensive processes, energy consumption, and associated environmental externalities, display features of rivalry and collective vulnerability that generate coordination problems analogous to those identified by Elinor Ostrom in CPRs.Footnote 259 In parallel, the epistemic layer of AI, including models and data, exhibits characteristics of a knowledge commons. This hybrid configuration calls for governance approaches that move beyond centralised, one-size-fits-all regulation, and that focus on value communication (even ‘cheap talk’) and trust to avoid resource overuse.Footnote 260
Within this framework, AI councils can be conceptualised as key institutional mechanisms for addressing CPR-like dilemmas in AI-driven energy systems, notably as trust accelerators, fostering repeated interaction, and early-warning systems for overconsumption, identifying emerging risks related to energy use, data extraction, and environmental impact before they crystallise into systemic failures. Crucially, councils would institutionalise deliberation on the long-term and intergenerational impacts of AI, both positive and negative,Footnote 261 embedding anticipatory governance capable of addressing the temporal asymmetries inherent in AI development.
A central mechanism would be the introduction of mandatory AI life-cycle disclosures, including energy consumption, carbon intensity, material dependencies, and data practices, ensuring accountability by rendering the material infrastructures of AI visible and contestable. In this way, the ward system provides the institutional infrastructure through which AI may be enacted as a form of distributed, reflexive, and ecologically bounded governance, aligning technological development with planetary limits while preserving democratic agency.
Beyond oversight, the ward system also enables the deployment of AI as a sustainability enabler, positioning citizens’ councils as co-designers of AI-enabled ecological transformation. In this capacity, councils would deliberate on and prioritise the use of AI in domains such as energy system optimisation, biodiversity monitoring, and climate risk management, thereby shaping socio-technical pathways aligned with sustainability goals. This participatory orientation reflects broader calls for mission-oriented and democratically guided innovation for planetary-scale computation,Footnote 262 moving beyond technocratic approaches.
Crucially, for such participation to be meaningful rather than symbolic, councils would need to be complemented by AI literacy hubs, designed to enable informed judgement. In Arendtian terms, the capacity for political action presupposes not expertise as such but the ability to form and express opinions in a shared world, thus engaging in political activity not as a means to an end but as an end in itself.Footnote 263 Literacy hubs would therefore function not as technocratic training centres but as enabling conditions for participation, bridging the gap between highly technical AI systems and the public reasoning required for their legitimate governance.
At the same time, the councils would be featured by power moving ‘neither from above nor from below’ but being ‘horizontally directed so that the federated units mutually check and control their powers’.Footnote 264 They could be imagined as a linchpin between public perceptions and debates and the institutions of liberal representative democracies. In Arendt, this federal network of councils would eventually produce parliamentary representatives,Footnote 265 for instance through bodies whose preventive advice should be strictly accounted for to pass AI regulation.
E. Fiduciary duties or rights?
A clarification is necessary regarding the relationship between the institutionalisation of fiduciary duties (second principle) and the ‘right to have rights’ (third principle), both grounded in the instrumentality of AI (first principle). The argument advanced in this article does not presuppose a direct horizontal application of human rights law to private AI providers or deployers. Such a claim would sit uneasily within the classical vertical structure of both EU fundamental rights law, anchored in Article 51(1) of the Charter of Fundamental Rights of the European Union, and the European Convention on Human Rights system, in which protection against private interference is mediated through positive obligations of the state rather than direct claims against private actors.Footnote 266
Instead, the proposed principles would operate through a mediated and institutionalised form of horizontality, characteristic of contemporary EU constitutionalism. The proposed diligentia diligentis obligations, as described earlier, are conceived of as legally structured duties of care embedded in contractual, regulatory, and compliance architectures governing AI development and deployment. Similarly, the Arendtian and more deliberative ‘right to have rights’ is understood not as a directly enforceable subjective entitlement against private actors but as a principle of institutionalised contestability that requires the legal order, primarily through the EU legislator, to construct and maintain spaces of meaningful deliberation, objection, and revision. This is particularly relevant in AI governance, where AI systems in energy governance are usually developed and deployed without meaningful ex ante public deliberation. This generates a structural asymmetry: individuals and affected communities are subjected to algorithmic forms of decision-making before having had any genuine opportunity to shape their design, objectives, or distributive consequences. The ‘right to have rights’, in this context, must therefore operate not only ex ante, through participation in design and regulatory processes, but also retrospectively, by reopening spaces for contestation, revision, auditability, and collective oversight of systems that are already operational.
On this reading, the proposed principles are rendered effective through their proceduralisation within EU law. Without such institutional mediation, rights risk remaining formally recognised yet practically unenforceable, a concern already implicit in Arendt’s critique of rights as dependent on membership in a political-legal order capable of guaranteeing their effectiveness.Footnote 267
VII. Conclusion
In this article, I have attempted to untether technological considerations from much of contemporary political philosophy.Footnote 268 Building on Arendt’s earthbound and constitutive legal perspective, I have argued that current AI regulatory frameworks are highly inadequate and destabilising for energy justice and political life. My central proposition lies in three constitutive principles for enabling a ‘justice-centred AI’ and recreating the common world that has been lost in the ever-growing alienation and mass society of energy consumers, starting from the European Union. To institutionalise such principles, I offer a framework where bonds and ties for the future are enshrined through institutional practice ‘without, however, destroying the fundamental contingency of action’.Footnote 269
More generally, as for other technological inflection points of the past (eg the printing press or the railways),Footnote 270 AI presents us with both a technical and a cultural shift. To overcome the outright commodification of data and other grave crises (eg climate change and energy transition consensus), current regulatory frameworks need to understand their intrinsically cultural, as opposed to normative, value.Footnote 271 In particular, the law can contribute to the much-needed cultural transformation of our relationship with AI. It can counter the doom-and-gloom narratives on AI takeovers of humanity, especially in the high-risk field of energy systems, by imposing non-voluntary obligations on AI providers and deployers and places of deliberation for non-expert stakeholders. Such findings open new avenues for future research on fiduciary obligations to be extended beyond corporate actors to include public authorities deploying AI systems, particularly where such systems affect access to essential services such as energy.Footnote 272
Overall, AI laws can contribute to displaying AI ‘as our own creative power’: belonging to humans by being human-made and human-chosen.Footnote 273 As a matter of fact, AI is thus neither artificial nor intelligent.Footnote 274 In contrast, AI systems mirror our own intelligence back to us.Footnote 275 Such backward knowledge has great potential but can also ‘magnify, occlude and distort what is captured in their frame’.Footnote 276 More dangerously, it can ‘induce a type of self-forgetting … that looses our grip on our own human agency and clouds our self-knowledge’,Footnote 277 preventing us from solving existential crises.
Conclusively, Arendt helps us reimagine our common world by entrusting technology with a mediating, rather than a purpose-led, function. Technology introduces scientific findings ‘into the everyday world of appearances and renders them accessible to common-sense experience’.Footnote 278 Conversely, questions raised by thinking—questions of meaning—are unanswerable by common senseFootnote 279 and, thus, by technology. By preserving the recognition of human thinking, willing, and judging through justice-regulated AI, with healthy optimism, we can consider that humans will continue to pose such unanswerable questions of meaning, remaining capable of new beginnings and a common world protected by the solid walls of laws.Footnote 280
Open access
