New Lancet Planetary Health paper calls for governing AI as a global commons to protect society and the planet

In the paper, Governing artificial intelligence for planetary health, Professor Creutzig and an international team of co-authors propose a structured framework that classifies AI risks into three interconnected domains. Social risks include misinformation, surveillance, labour displacement, and the concentration of power among a small number of technology firms. Planetary risks arise from the rapidly growing energy, water, and material demands of AI infrastructure, including data centres and chip production. Safety risks stem from system failures, military applications, and the potential loss of human control over increasingly autonomous, agentic AI systems. 

“AI is already reshaping social systems and planetary dynamics,” “If governance continues to focus narrowly on technical performance or speculative future risks, we will miss the immediate and cumulative impacts on inequality, democratic institutions, and the Earth’s life-support systems”. 

Professor Felix Creutzig, Bennett Institute for Innovation and Policy Acceleration, University of Sussex

The authors identify data, energy, and compute as the core regulatory levers that cut across all three domains. They argue that regulating compute – such as monitoring and limiting access to large-scale computing infrastructure – can slow unsafe AI scaling and help retain meaningful human control. Data regulation, including audits and limits on data appropriation, is presented as essential to reducing bias, surveillance, and power concentration. In the planetary domain, the paper highlights the climate and resource impacts of AI, noting that data centres already account for low single-digit shares of global electricity use and could rise sharply as AI deployment accelerates, particularly where energy is fossil-based.

A central policy recommendation is to explicitly limit the development and deployment of high-risk agentic AI systems – systems capable of autonomous, goal-directed action – while incentivising non-agentic AI that supports human decision-making. The authors also warn that efficiency gains from AI risk being offset by rebound effects, where lower costs drive higher overall consumption, increasing energy use and emissions rather than reducing them. 

“Treating AI governance as a global commons means recognising that unregulated competition for scale and dominance is itself a root cause of risk,” Creutzig said. “We need international coordination that aligns AI development with planetary boundaries, social well-being, and the preservation of human agency”.

The authors conclude that without deliberate intervention, current AI business models will continue to amplify social polarisation, environmental degradation, and safety hazards – undermining both planetary health and democratic capacity to respond.