Artificial intelligence is often framed through acceleration: faster workflows, lighter models, greater efficiency. Yet efficiency does not automatically lead to sustainability. As systems become easier to use, they tend to be used more. This is the quiet tension behind Jevons’ Paradox: lower emissions per task can still produce a growing environmental footprint overall.

From this tension, the CO₂ Tracker takes shape. It makes the environmental cost of AI visible by estimating emissions from text prompts, token processing, and AI-generated images. Instead of presenting numbers in isolation, it translates them into visual patterns, cumulative histories, and everyday equivalents. A single interaction becomes part of a larger picture of use.

Grounded in AI Horizon’s Responsible AI Framework, the project treats sustainability not as an add-on, but as part of responsible system design: ethically justified, legally aware, and technically precise. Principles such as intergenerational justice, ecological integrity, transparency, and accountability shape how impact is measured, explained, and communicated.

In collaboration with LMU Munich, the CO₂ Tracker is also being examined empirically. At the center of the study is a simple but important question: does visibility change behavior? When users can see the footprint of their AI interactions, do they ask differently, avoid repetition, or develop a more deliberate relationship with the technology?

As a first step toward a broader culture of sustainable AI, the CO₂ Tracker brings environmental impact out of the infrastructure and into view: measurable, understandable, and part of how AI systems are evaluated.