Introduction

Artificial intelligence (AI) development is improving at an extraordinary speed, demonstrating reasoning capabilities once thought uniquely human. Behind these systems lies a vast infrastructure of data centres and semiconductors deeply entangled with environmental degradation and social inequality. AI has transformative potential to accelerate sustainability solutions or amplify existing risks. Governance is positioned as the foundational mechanism through which society can shape this potential to either support sustainable development or undermine it.

Artificial intelligence

AI matches or exceeds human capabilities in discovery, reasoning, and decision-making. It is categorised as Narrow AI (ANI), which is the only type existing today, Artificial General Intelligence (AGI), which remains hypothetical, and Artificial Superintelligence (ASI). ANI includes generative AI and agentic AI capable of autonomous planning. The broader ecosystem includes various data modalities such as text, image, audio, and spatial data, often processed within Large Language Models (LLMs) or multimodal foundation models.

Artificial intelligence and sustainability transitions

The current mode of development degrades the Earth’s finite capacity. Environmental crises are compounded by growing inequality, with 3.5 billion people remaining poor (44% of the global population) and 700 million living in extreme poverty (8.5%). AI can raise productivity, improve market matching, and support circular economy models. However, without deliberate guidance, these technologies risk amplifying existing inequalities, concentrating power, and placing strain on social and institutional systems.

Planetary environment

Opportunities include improved monitoring of Earth systems using remote sensing and image recognition to map biodiversity and detect deforestation. AI can refine scenario planning and predictive forecasting, enabling proactive disaster risk management. However, risks are significant; the entire AI value chain puts intense pressure on natural resources. High electricity demand for data centres contributes to CO2 emissions, while a surge in electronic waste outpaces recycling capacities, leading to pollution from toxic substances like lead and mercury.

Energy systems

AI can make energy systems more efficient by optimising resource use and reducing waste. It supports the integration of renewables and grid modernisation through improved forecasting for power demand and supply. AI may optimise system operations and maintenance to increase electricity transmission capacity without building new lines. Risks include fossil fuel entrenchment if AI enhances hydrocarbon recovery. Furthermore, the Jevons Paradox suggests efficiency gains may offset expected reductions in energy consumption if the additional demand remains fossil-fuel based.

Industry and labour

AI acts as a general-purpose technology reshaping industrial structures and drives productivity through task reallocation. Globally, approximately 24% of jobs have exposure to generative AI, rising to nearly one in three jobs in high-income countries. Automation risks displacement for routine tasks, while benefits may concentrate among high-skilled workers. Additional risks include increased market concentration among a few large technology firms and workplace alienation caused by algorithmic management. Estimates suggest AI could create as many as 170 million new jobs, yet 39% of executives have already reported making headcount reductions in anticipation of AI-driven efficiencies.

Finance

AI improves financial efficiency through enhanced risk assessment, portfolio optimisation, and improved data availability for emerging markets and developing economies. However, risks include financial instability and market concentration. The capital-intensive model driving AI expansion may create vulnerabilities, such as “roundabouting” (circular financing) within the data centre sector. Poor-quality data or flawed algorithms can also undermine sustainability reporting and amplify existing biases in credit assessments.

Democracy and societal resilience

AI offers inclusive civic participation by translating and summarising complex policy content. Governments are using AI to analyse citizen input and moderate online participation. Conversely, risks include the proliferation of disinformation, trust erosion, and the potential for “black box” models to cause a loss of human control. Existential threats are noted if AI develops capabilities that match or exceed human intelligence without adequate oversight. Control is currently concentrated among a few firms and countries, raising concerns about political influence.

Summary of opportunities and risks

AI is rapidly reshaping societal management of the environment, energy, and finance. It can support sustainability transitions only if deliberately governed and aligned with public goals. Sectoral risks are often cross-cutting, stemming from AI’s scale, speed of diffusion, and resource intensity. Addressing these risks requires both sector-specific safeguards and coordinated, cross-cutting governance responses to manage market concentration and human oversight.

Sectoral and cross-cutting perspectives

Addressing challenges requires tailored sectoral governance and cross-cutting safeguards. Recommendations include implementing “human-in-the-loop” processes, regular auditing for bias, and the creation of dedicated AI oversight bodies. For the planetary environment, policies should require data centre developers to adopt holistic designs to minimise footprints. In energy, governments should increase public R&D investments in green-AI hardware and software. Labour policies must prioritise human-AI complementarity and invest in foundational digital literacy.

A global governance framework for AI

The report proposes a phased roadmap. The Short Term (0–2 years) focuses on building scientific infrastructure, including the operationalisation of the UN AI Scientific Panel and an AI Expert Technical Laboratory. The Medium Term (2–4 years) involves an interim safety framework focused on harm prevention, potentially including temporary restrictions on high-risk research. The Long Term (4–6 years) envisions a binding but adaptable global framework convention on AI to establish shared obligations and accountability mechanisms.

Conclusion

AI is neither an inherent saviour nor an inevitable threat. Its impact on sustainability transitions will be dictated by deliberate choices. Governance is the central lever for aligning AI with long-term societal and planetary goals to ensure it acts as a force for good, anchored in safety and accountability.