Introduction

1On November 30^th 2022, OpenAI announced the release of ChatGPT, a “model […] which interacts in a conversational way.”¹ The declared purpose of the launch was “to get users’ feedback and learn about its strengths and weaknesses.” Since then, the model has undergone extensive reviews and frequent updates and it has triggered a race towards AI supremacy that is reshaping geopolitics, public and private investments and technology itself.² As model performance converges overtime, competition at the technological frontier of AI is becoming increasingly tight.³ Technical superiority alone is unlikely to determine the future landscape of winners and losers. Instead, success will hinge on other critical factors. Foremost, among them, alignment with human values and community-shared principles will become key. Technology-related risks are perceived as key short-term threats by the general public,⁴ with AI diffusion strongly linked to societal fragmentation and conflicts.⁴ These concerns come at times of growing AI ecological footprint, with preoccupations linked to the use of natural resources and a growing energy demand to power AI infrastructure, model training and inference.⁵ However, while the ethical risks of and implications from AI have received considerable attention from scholars,⁶ policy-makers⁷ and civil society organizations,⁸ the governance aspects of planetary stewardship of AI remain relatively absent from the public debate. However, the climate transition to a low-carbon and more resource efficient economy and the digital revolution are far from being disconnected. Instead, they are linked under the framework of a “twin transition” in which technology can accelerate and streamline activities with implications for both climate mitigation (i.e., reduction of the emissions) and adaptation (i.e., management of the impacts).

What is so Special About AI?

2Although the current AI hype may seem unprecedented in the eyes of the public opinion, the AI summer unfolding today is the product of a much longer and complex chain of innovations. From a systemic and socio-economic perspective, the ability to master knowledge and its diffusion processes lie at the heart of technological progress. The telephone (1876-1877), the first device ever invented to voice two distant agents in real-time, enabled a profound cultural shift: it is responsible for significant changes in the workforce, including a stronger female participation, especially during war times. This device is the initiator of the connected economy, which ultimately led to the mass-deployment of the internet and the fast diffusion of broadband networks in the 1980s. At the core of the connected economy lies the capacity to communicate and share data, information, and knowledge generated remotely from the recipient, in real time. Regulation, particularly in the European Union (EU), has focused extensively on competition law to ensure fair access to infrastructure for incumbents and to promote transparency. Notable examples include the Open Network Provision in the EU (Directive 97/51/EC) and, in the United States, the Cable Communications Policy Act of 1984.

3The connected economy emerged to meet the growing need of global exchange and trade freedom, enabling the movement of data and information that powers the knowledge hubs of the present and future. This seamless flow of information fuelled business models innovation, with digital platforms increasingly delivering goods and services and marking the beginning of the digital economy era in the early 2000s. Although traditional production and consumption models largely remained unchanged, digital tools allowed products and services to cross borders with ease, fostering a truly global market. This shift also gave rise to information-based services that support decision-making at both corporate and governmental levels. Governance mechanisms in the United States (US) and the EU began to evolve in response to the challenges of the digital age, particularly around privacy and harmful content. In the EU, the Directive on Privacy and Electronic Communications (Directive 2002/58/EC)⁹ was introduced to safeguard the confidentiality of online communications and regulate unsolicited messaging (e.g., spam). In the US, regulations such as the Electronic Signatures in Global and National Commerce Act (2000)¹⁰ established the legal validity of electronically signed contracts in both interstate and international commerce, enabling new business models to comply with legal requirements. These regulatory developments marked the emergence of new digital rights and a reconfiguration of economic processes, positioning data and information as valuable commodities and as new objects of exchange in the digital era.

4The machine-enabled intelligence transformation ‒ of which AI is a core component ‒ marks a shift in the role of knowledge from being merely the object of exchange to becoming the subject of it. Knowledge no longer functions solely as a tool for decision-making; it emerges as an indispensable asset across multiple value chains. This shift signals the onset of the digitalized economy, characterized by two defining features. First, knowledge represents the primary driver of value creation. This is well exemplified by the capabilities of generative AI in material science and scientific discovery. AI holds the potential to reduce or even eliminate dependency from natural resources by introducing laboratory-created input materials into the energy transition process.¹¹ Intelligent algorithms also support energy grid management by optimizing workloads and enabling more diversified energy production ‒ contributing to the increasing integration of renewable energy sources and to meeting the decarbonization goals to keep the temperature within the 1.5 degrees target set a decade ago in Paris. Second, the distinction between users and producers of data and information is increasingly blurred. Generative AI systems continuously evolve through user interactions, integrating feedback to refine their outputs. However, these outputs are often absorbed seamlessly into daily life, typically without external verification or fact-checking ‒ raising concerns about the reliability and traceability of the information they produce.

5Traditionally, regulation and policy have focused on competition and data protection. However, in the age of AI, immateriality of operations, blurred and fine lines between owners and beneficiaries make existing policy tools at risk of obsolescence. As challenges become more intertwined and complex, AI governance must evolve to address high-level goals and ambitions for society at large. Inspired by the mission-oriented approach embraced by the European Commission,¹² the new governance of AI must set clear and ambitious goals operationalized by shared criteria. A compelling illustration of this new governance trajectory is the notion of planetary stewardship, with which the trajectory of AI development must become increasingly aligned.

The Earth Alignment Principle

6The concept of “alignment” ‒ ensuring that AI systems reliably follow human-intended goals and values ‒ was historically proposed by academic and industry actors to address three challenges:

1. design agents that surpass human capabilities (“smarter-than-human”) while reliably pursuing their specified objectives;

2. rigorously define and unambiguously communicate the goals assigned to them;

3. ensure that, despite inevitable mistakes in their initial programming, these agents remain cooperative with their “creators.”¹³

7Governance of AI systems typically addresses misalignment retrospectively via the collection of evidence and the implementation of corrective measures.⁶ Ethical principles are intended to guide systems’ training in a forward-looking manner. Research in ethics, philosophy and socio-technical systems has largely focused on alignment as a means to prevent AI from causing new or exacerbating existential or high-impact risks, and to proactively mitigate harms generated by misaligned systems.¹⁵ While these efforts have laid important foundations for ex-post corrections, there is a growing recognition that a truly aligned-by-design AI can only emerge from the application of solid, shared, and high-level principles.

8As climate change impacts become more frequent and severe, the need to promote a just, ordered and scalable transition to a low-carbon and more resource efficient economy, while also establishing context-relevant adaptation measures is urgent.

9In 2025 a heterogeneous group of authors introduced the Earth alignment principle for AI development which not only serves a risk-mitigation purpose, but also suggests how to redirect AI growth and diffusion towards planet-positive applications to enhance people’s welfare and to reduce as much as possible the acceleration towards Earth’s destabilization.¹⁴ This proactive intention to leverage AI’s capabilities will promote applications that range from weakly to strongly aligned over the course of their entire life cycle. As by the mission-oriented approach, the Earth alignment principle unfolds through three concrete and operational criteria for development, deployment and use. While each of these criteria, Earth alignment is not binary. AI systems will vary in level of alignment from strong to weak or even strong misalignment.

10Three criteria for strong Earth alignment¹⁴:

1. AI systems should help to accelerate the transition to sustainable production and consumption in ways that respect planetary boundaries, or at least do not obstruct these objectives

2. AI systems should be developed, deployed and used in ways that ensure equitable access to AI tools for global sustainability and avoid concentrations of power

3. AI systems should be developed, deployed and used to support greater societal cohesion, build trust and provide access to reliable information for planetary stewardship.

The Three Criteria in Practice

11The first criterion actively prioritizes some applications over others to address the climate and biodiversity crises with decarbonization and nature preservation at heart. The electrification of our economies is driving the transformation of the energy sector worldwide. Companies increasingly deploy AI to optimize energy and resource supply and consumption, as well as electricity transmission and balance.¹⁵ AI applications can address energy loss levels especially in developing countries and it can be used to discover new and more efficient materials for photovoltaic systems.¹⁶ As AI’s energy footprint is far from negligible, but still contested in magnitude,¹⁷ ex-ante and transparent impact assessment, disclosure and reporting are needed to mitigate the emission profile of training and inference. According to the IEA, the global electricity demand from data centers has grown at 12 percent annually over the past five years,¹⁷ but the demand pull in data center intensive locations can reach up to 20 percent (e.g., Ireland) and 25 percent (e.g., Virginia). Data centers for AI are projected to emit between 24 to 44 Mt CO₂ equivalent per year between 2024 and 2030 in the US alone.¹⁸ The Earth alignment principle does not neglect these negative impacts and its first criterion encourages AI applications which support carbon reduction to achieve a net positive balance in the future. Beyond energy, AI applications can support biodiversity monitoring and land-use planning which identify and solve social, ethical and governance challenges that underpin the twin transition. Recent assessments of the AI expansion^{19, 20} highlight that the use of AI within the biodiversity context enables locally relevant decisions by offering highly granular insights and empowers those very same communities which do not have reliable access to information.

12The second criterion addresses the growing concern of unequal and inequitable access to AI systems, models and infrastructure. The unprecedented adoption speed of AI technology (Figure 1) reflects the extent to which this general-purpose technology is permeating multiple sectors of the economy. AI use, already polarized and highly concentrated by geography (Figure 2a), is even more strongly dominated by a few a countries when used for sustainability²¹ (Figure 2b). The twin transition may widen the digital and climate preparedness divide, exposing those who invest less to even bigger risks. The second criterion addresses this problematic diffusion and encourages technology and infrastructure transfer, as well as capacity building to ensure benefits are shared equally. For example, global initiatives such as the AI Climate Institute (launched at COP30 in Belém, Brazil) aim to support shared research capacity and knowledge exchange across regions.²²

13As the most affected by climate impacts are developing nations who contributed in very limited measure to greenhouse gas emissions, the application of AI technologies can mitigate the negative effects, improve planning and support rapid shifts towards clean technologies. The AI Skills coalition by the ITU AI for Good Initiative is a successful example of private-public partnership to build AI skills across the world. Particular attention is devoted to the public sector and administration, where data governance and responsible AI principles can embrace the Earth alignment principle and the second criterion with stronger emphasis.

14The third criterion involves the synergic interaction between social, natural and technological factors. A new governance of the commons (including the digital ones) can rebuild institutional trust. By proactively redirecting AI deployment towards planetary stewardship, societies can reestablish trust in the democratic process and shared principles. Operationally, this goal can be achieved in two ways. First, in the age of never-ending information, the promise of interconnected societies is not delivering cohesive societies, but divided and polarized positions. By acting and sense-making technology, AI can find and synthesize relevant information to enable decisions and maximize impact. The World Bank ImpactAI tool is an example of such an effort and informs responsible investment strategies to close the financing gap of the United Nations Sustainable Development Goals (SDGs), including SDG13 (Climate Action), SDG14 (Life Below Water) and SDG15 (Life on Land). Fine-tuned AI models are also helpful in detecting greenwashing and false or inaccurate claims in climate and nature-related risk reports.²⁴ By exposing corporate malpractices, AI facilitates consensus over the Green Premium (i.e., the additional markup to greener alternative than conventional industrial products).²⁵ The second channel involves the co-creation of AI tools with relevant stakeholders, fostering literacy, ownership and control of models and platforms. Collective intelligence may entail specific steps in the knowledge production chain (i.e., data collection, data validation, etc.) or co-production of the whole product.²⁶ Moving beyond the “human-in-the-loop”, citizens, companies and organizations can work together to collect relevant data.

15Although specific policy measures must be tailored to local contexts and designed to be fit for purpose ‒ and will therefore differ across regions ‒ the underlying guiding and governance criteria should be universally applicable.

Conclusion

16The Earth alignment principle emphasises that no criterion should be developed in isolation; each must be designed and implemented in coordination with the others.

17From a governance perspective, this is reflected in redistributive policies that both reduce emissions and promote fairness, thereby driving systemic change. In the transport sector, for example, governments can introduce income-dependent incentives that encourage people to shift from private cars to public mobility options, tailoring measures to commuters’ preferences and constraints. AI tools can support this by analyzing high-quality, diverse data on mobility habits, while ensuring that anonymity and privacy are safeguarded. AI can also enhance the operations of high-emitting sectors by lowering fixed costs and improving efficiency. Under the Earth alignment principle, good governance does not undermine these benefits, but seeks to channel part of the resulting gains into compensation and support mechanisms that avoid harm to the environment and communities, and ideally contribute to their well-being.

18For the Earth alignment principle to be effective, it must be endorsed and accepted by the international community. Because the twin transition unfolds at a global scale, yet generates local impacts and depends on interconnected, cross-border knowledge networks, robust multilateral arrangements are the most appropriate means of building shared agreement on both the rationale for, and the implementation of, the principle. Although specific policy measures must be tailored to local contexts and designed to be fit for purpose ‒ and will therefore differ across regions ‒ the underlying guiding and governance criteria should be universally applicable.

19As more capable models change how we consume and produce, our societies and their institutions may shift into unforeseen configurations that risk accelerating the depletion of ecosystems and further destabilizing the planet. While AI shares many of the challenges associated with earlier technological innovations, it also introduces distinct novelties: in how intellectual property is protected, in its largely immaterial yet ubiquitous footprint, and in the hybrid public-private structure of its development. From a governance perspective, these features call for changes in how we conceive of and design policy. Rather than relying mainly on reactive, corrective measures, proactive and forward-looking, mission-oriented governance has the potential to steer the growth and deployment of AI towards the common good. The Earth alignment principle provides one such example, addressing the urgent needs of the future: mitigating and adapting to climate change. As a framework for coordinated governance, it can enable AI to accelerate a planet-positive transition without triggering new instabilities.