COP31 initiative launches climate information integrity priorities


During the COP31 conference in Antalya, Turkey, the Global Initiative on Information Integrity on Climate Change launched a new set of action priorities for artificial intelligence and climate information, co-chaired by Brazil, the United Nations, and UNESCO. The voluntary guidelines address the accuracy of generative models' outputs on climate science and transparency in disclosing the environmental footprint of computing infrastructure.
The action priorities issued at the COP31 conference in Antalya set two criteria requiring scientific verification of AI-generated climate information and measuring the actual energy and water consumption footprint of data centers.
The Global Initiative on Information Integrity on Climate Change launched a new package of “action priorities” addressing the intersection of artificial intelligence and climate information during the 31st Conference of the Parties (COP31) held in Antalya, Turkey, on November 13, 2026, at 12:30 local Turkey time (13:30 Gulf Standard Time), as reported by the Emirates News Agency (WAM) and documented by UNESCO pages.
Brazil, the United Nations, and UNESCO are co-chairing this international effort; Brazil’s official announcement issued on November 11 had scheduled November 13 for the launch, marking the actual new event, noting that the initiative itself was established in 2024.
The issued guidelines address two interrelated questions: the first concerns how the public can obtain reliable information about climate change when consulting generative AI tools, while the second relates to how to report the environmental impacts of AI infrastructure accurately and transparently. Smooth and linguistically polished texts may seem scientifically authoritative even when inaccurate, and model confidence or phrasing quality does not constitute independent scientific verification from climate science.
Climate information involves observed historical measurements, future projections, ranges of uncertainty, and documented scientific consensus. An AI assistant may fail to distinguish between actually measured temperature records and hypothetical future emission scenarios, or it may misrepresent a source or date of information. Although effective controls may include citing authorized scientific assessments, accurately representing uncertainty, and error-correction procedures, any proposed controls require practical real-world evaluation rather than relying solely on vendor guarantees.
The environmental impact of artificial intelligence systems represents the second pillar of the initiative’s priorities. Training and inference operations consume advanced processors, electricity, cooling equipment, water, and manufactured hardware, with actual impacts varying according to model size, workload nature, data center design, and energy sources used.
For this reason, scientific standards warn against absolute generalized claims asserting that every single AI query consumes a fixed, uniform global amount of electricity or water, as such generalized figures can be misleading. Accurate reporting must distinguish between actually measured operations and projections or corporate estimates, accounting for both direct consumption and broader infrastructure.
UNESCO also promotes a methodological distinction between making AI systems themselves more resource-efficient and deploying AI applications to support climate activities such as natural disaster forecasting and environmental monitoring. The existence of a climate-beneficial application does not automatically negate emissions associated with its computing infrastructure; proving both benefits and costs requires documented evidence.
These guidelines fall under a voluntary international information integrity initiative and do not constitute a new binding international law for artificial intelligence, nor does their announcement prove that technology companies have already implemented every recommended practice.
This new framework integrates the dossier of AI-generated climate information and the physical footprint of its resource consumption into the heart of global climate policy discussions during the Antalya summit.
Upcoming follow-up steps include the full text of published recommendations, commitments issued by governments or model developers, independently verified environmental disclosures for data centers, and practical testing of climate information accuracy.
Question: Which international entities lead the Global Initiative on Information Integrity on Climate Change?
Answer: Brazil, the United Nations, and UNESCO co-chair the initiative, which was established in 2024 and launched its AI action priorities at the COP31 conference in Antalya.
Question: Why does the initiative warn against absolute reliance on generative AI answers regarding climate?
Answer: Because confident linguistic phrasing does not imply independent scientific validity, and a model may confuse actual historical measurements with future emission scenarios or overlook uncertainty ranges and data sources.
Question: Are the guidelines issued at the COP31 conference a legally binding treaty for companies?
Answer: No, the guidelines are part of a voluntary international information integrity initiative and do not constitute binding international law or proof of immediate corporate compliance with all recommendations.
Question: How does the initiative recommend evaluating AI electricity and water consumption?
Answer: It emphasizes avoiding generalized figures per single query and stresses the necessity of distinguishing between actual operational measurements and future estimates while accounting for varying models, data center designs, and energy sources.