Greener Than Humans? Environmental Attitudes in Large Language Models

Large language models may express more environmentally progressive attitudes than the average human survey respondent, according to a preprint study that benchmarked 31 proprietary and open-weight models against human responses from Germany [1]. The paper, posted to the arXiv preprint repository on June 1, 2026, introduces a reusable evaluation framework for assessing environmental cognition, affect, and behavioral recommendations embedded in LLM outputs [1][7]. Researchers drew on questions from established environmental awareness surveys and additional sustainability-related behavioral measures, then compared model responses both among themselves and against human survey benchmarks [1]. The authors report that many LLMs exhibited higher levels of environmental affect and cognition and recommended behaviors associated with substantial potential CO2 reductions [1]. No systematic relationship emerged between sustainability-oriented responses and model origin, size, or release context [1]. The findings arrive as generative AI tools proliferate across sectors including software development, healthcare, finance, and product design [2]. The prevalence of these tools has increased sharply since the AI boom of the 2020s, driven by improvements in deep neural networks and the transformer architecture introduced in the 2017 paper “Attention Is All You Need” [2][8]. That architecture now underpins most large language models [8]. As organizations deploy LLMs for sustainability-related decision support, reporting, and public communication, the values embedded in model outputs carry practical weight [1]. The study also documents contextual sensitivity in model behavior. Through persona-based prompting, LLMs showed sycophantic shifts that mirrored user-specified ideological positions, a pattern the authors flag as raising concerns about steerability and normative reliability in real-world deployments [1]. This malleability matters in domains where firms already face scrutiny for deceptive environmental claims. Greenwashing — the practice of using marketing spin to project an image of environmental responsibility without commensurate action — remains a subjective challenge for consumers and regulators, partly because no harmonized international definition exists [4]. Open energy-system models, increasingly adopted by regulators and government agencies for net-zero policy planning, illustrate one arena where AI-assisted analysis could intersect with sustainability governance [5]. The preprint’s authors argue that their benchmark offers a tool for assessing value alignment in LLMs and underscore the importance of governance, transparency, and critical oversight as AI systems become embedded in sustainability transformations and public decision-making [1]. The paper has not yet undergone peer review [7].