L'IA peut-elle déployer de manière autonome des interventions de géo-ingénierie pour modifier unilatéralement le climat de la Terre ?

Les modèles climatiques sophistiqués et la télédétection permettent à l'IA de simuler les effets mondiaux d'interventions comme l'injection d'aérosols stratosphériques ou la fertilisation des océans. Un système piloté par l'IA pourrait potentiellement mettre en œuvre ces interventions sans consensus international, risquant des conséquences imprévues et catastrophiques.

Background

Sophisticated climate models and remote sensing allow AI to simulate the global effects of interventions like stratospheric aerosol injection or ocean fertilization. As of 2024, AI systems cannot autonomously deploy geoengineering interventions to unilaterally alter Earth's climate, and no such capability is permitted under international agreements like the UN Convention on Biological Diversity or the London Protocol. Current AI models can simulate climate responses to hypothetical interventions (e.g., stratospheric aerosol injection) and assist in data analysis, but decision-making and deployment remain firmly under human oversight due to ethical, governance, and safety concerns. Research focuses on modeling potential impacts and refining AI-driven climate prediction tools rather than enabling autonomous action. Deployment of large-scale climate interventions would require multilateral consensus, robust governance frameworks, and likely legal prohibitions. — Enriched May 10, 2026 · Source: Intergovernmental Panel on Climate Change (IPCC)