Sasha Luccioni, a former sustainable AI researcher at Hugging Face, has left the company and started Sustainable AI Group, with a clear target: getting products such as ChatGPT and Claude to display energy use and carbon emissions next to each user query. Her argument is simple. People are using generative AI without seeing the environmental cost tied to each interaction, and that lack of disclosure leaves companies with little basis to judge the tradeoffs.
Companies are starting to ask what AI use means for ESG targets
In an interview with WIRED, Luccioni said she is hearing a growing question inside large organizations: if management requires employees to use tools like Copilot, what does that mean for corporate ESG goals? There is no standard answer yet. Major AI providers do not show per-query energy or carbon data in their interfaces, so users of ChatGPT or Claude are given no direct signal about environmental cost.
She wants those figures placed directly in the product interface. In her view, this is not only about transparency. It is also a competitive issue. An AI company that adopts renewable-powered data centers and openly reports consumption data could gain a market edge. Regulatory pressure is building as well. The EU AI Act includes sustainability provisions, and the first reporting obligations are starting to take effect. In Asia, countries including those working with the International Energy Agency are also moving toward stronger data-center transparency requirements.
AIEnergyScore exists, but major model providers stayed out
While at Hugging Face, Luccioni built AIEnergyScore, an open-source ranking system designed to compare the energy efficiency of AI models under a common benchmark. The problem is participation. The system can only measure models whose developers submit data, and major providers including OpenAI, Google, and Anthropic have not done so.
Luccioni sees a structural conflict in that refusal. Large AI companies often sell both model access and the underlying compute. The larger and more expensive the model a customer uses, the more compute revenue can follow. Under that setup, steering customers toward smaller and more efficient models can run against revenue incentives.
Task-specific models can use a fraction of the compute
She also points to data she has tracked over time showing that, across many enterprise use cases, much of the actual productive AI work has long been handled by classifiers trained for specific tasks rather than general-purpose large language models. For a task like identifying the sentiment of customer service emails, a dedicated classifier may require less than 1% of the compute needed by a GPT-4-level model. Without comparison data, companies may be choosing models 10 times or even 100 times larger than what the task really requires.
Her position is not anti-AI. The core claim is that matching model size to task complexity is an engineering and procurement decision that can reduce both spending and emissions.
Sustainable AI Group targets procurement and compliance pressure
After leaving Hugging Face, Luccioni co-founded Sustainable AI Group with Boris Gamazaychikov, formerly Salesforce’s chief sustainability officer. The pairing is deliberate: one side brings technical measurement, the other brings enterprise sustainability governance. The shared goal is to turn sustainability demands into practical standards for buying and deploying AI tools.
Luccioni argues that the EU timeline may turn this from an ethical debate into a compliance issue. Once companies are required to report sustainability data under the AI Act, a supplier’s refusal to provide energy-use information becomes more than a transparency gap. It can become a compliance risk. Her new group aims to support external validation around that decision framework while continuing to pressure AI companies to disclose energy data.