AI transforms urban planning but poses critical governance risks, Nature review warns

Artificial intelligence is fundamentally reshaping how cities are designed and managed, though its integration into critical infrastructure introduces significant safety and equity risks. A new review published in Nature by Northeastern University professors Esteban Moro and Ryan Wang analyzes more than 100 studies to determine how AI is currently being deployed in urban science.

The research details a shift toward geospatial AI, which merges machine learning with satellite imagery and mapping. Urban planners are now utilizing distribution-fitting generative models to simulate disaster scenarios; for example, by training on global flooding imagery, these models can predict specific flood impacts in areas like Boston's city center. Beyond physical simulations, foundation models—specifically Large Language Models (LLMs)—are being used to translate dense, technical policy documents into plain English to make urban planning more accessible to the general public. In Beijing, the researchers found that AI agents used to represent resident perspectives during planning meetings actually improved inclusion and participant satisfaction.

The Rise of Geospatial AI

This transition is driven by the ability of AI to handle complex simulations that were previously impossible for human planners to map in real-time. By synthesizing vast amounts of geospatial data, cities can move from static zoning maps to dynamic models that predict how a new building or a climate event will affect the surrounding environment. However, this capability is often obscured by the "black box" nature of AI decision-making, where the logic behind a specific urban recommendation is not transparent to the officials implementing it.

Governance and Safety Risks

Despite the efficiency gains, Moro and Wang warn that relying on AI for high-stakes infrastructure can be dangerous. Inaccurate outputs in disaster response or large-scale construction can lead to physical danger and systemic governance failures. "In high-stakes settings such as disaster response or large-scale infrastructure planning, these limitations become governance risks," the researchers noted. The authors argue that the industry must stop treating AI as a "toy" and instead implement rigorous validation systems and "human-in-the-loop" reviews to ensure that AI-generated plans are safe and equitable.

A Framework for Ethical Planning

To mitigate these risks, the researchers propose a formal ethical framework for urban AI centered on five core themes: fairness, privacy, explainability, controllability, and community participation. This framework aims to prevent the amplification of existing urban biases and ensure that AI tools remain under human oversight. As Ryan Wang emphasized, the priority must be honest evaluation over novelty: "While we use it it’s important to evaluate it honestly. We can’t just say there’s a new toy. Let’s use it."