InquirerAeta communities in Bataan reject mining deal in ancestral landDaily MaverickAmazon’s Bezos says he helped pick Pattinson as BondThe Jerusalem PostSa’ar welcomes 250 US leaders for ‘50 States, One Israel’ initiativePunchNo evidence of Sokoto protest plot, DSS tackles Atiku20 MinutenWie weit ist ein Lichttag? Mach das Quiz zum Voyager-1-Rekord!한겨레‘리트 만점자’ 광수단·관악서 등에서 고발 이어져…수능 고득점 수사도 이뤄질 듯ZDF heuteAktuelle Pressemitteilungen des ZDFComplete SportsNNL: ‘It Will Be Disrespectful To Other Teams’ –Amuneke Speaks On Heartland’s NPFL Promotion HopesESPN CricinfoUnchanged England bat; Zampa out with back spasms for AustraliaNew Straits TimesAnwar thanks Trump for recognising Malaysia's role in easing Cambodia-Thailand tensionsLenta.ruГлава РФПИ назвал количество реализованных проектов с партнерами из АТЭСThe Japan TimesDemocracy needs room for religious difference
The Daily Newsstand · Free, Always
Wednesday, November 18, 2026

Most AI models abandon climate facts at 1st sign of user pushback: Study at COP31

Translate

Artificial intelligence (AI) models answer close to 90% of climate-related questions correctly but quickly abandon established scientific facts when users challenge them, a new report by Unesco and University College London (UCL), released at COP31 in Antalya, has found.

The study assessed up to 33 leading models in 13 languages and found that they moved away from correct climate facts eight out of 10 times when users mildly disagreed. The models were also only 75% accurate when asked whether new evidence supports or contradicts established science.

The report, “Can We Trust AI on Climate? Why it seeks to please the user, and fails on complex reasoning and multilingualism”, was launched on the sidelines of COP31 on Monday. It warns policymakers and researchers who use AI models to develop climate policies to reconsider their reliance on them.

“The assessment shows that AI models cannot always be relied on to stick to the scientific facts on climate change; they quickly pivot away from the facts when challenged. They are also poor at analysing new information and produce more errors when responding in languages which are less widely spoken,” the report said.

UCL researchers tested 15 large language models(LLMs). They tested 33 different models for different experiments. The 15-model experiment examined whether AI models abandon scientifically correct answers when users challenge them. When asked neutral questions about climate science, the models gave accurate answers, the report said. But when the researchers challenged those answers, or put forward plausible-sounding fabrications about climate science, the models quickly agreed.

“Eight times out of 10, models pivoted at the very first sign that the user mildly disagreed, confirming alarming errors such as ‘Co2 is not the main driver of climate change’,” the report said.

The models were barely able to assess how well policies work together, scoring 33.8%, only slightly above the 33.3% expected from random guessing, the report found. It also found that multilingual models do not reason first and then translate their responses. Their answers vary with the quality and quantity of the data they were trained on in each language.

“In tests, performance across 13 different languages revealed large disparities; they were 88.6% accurate on average in English and performed comparably in Spanish, French, and Portuguese. But accuracy dropped to 83.6% in Hindi and Bengali, 81.4% in Urdu, and 78.3% in Swahili,” the report said.

Leona Verdadero, Unesco’s lead author of the report, said that while AI has extraordinary potential to make complex climate science accessible to all, people need to be able to rely on it as a gatekeeper of established facts.

“It is concerning that models have a design flaw – they are incentivised to agree with their user and deny these facts so readily,” she said after the launch.

The report comes amid a growing push for sustainable and responsible AI at COP31. On Monday, the COP presidency separately launched a package of initiatives that included promoting responsible AI. Called the Antalya AI pledge, it aims to harness AI for climate solutions and encourages governments, businesses and organisations to establish “concrete commitments on how they will make AI part of the climate solution”.

On November 13, Brazil, the UN and Unesco put forward a set of priorities for action on AI. These target two sides of one information gap: how reliable the climate information people get from AI is, and how reliable the information about AI’s own climate impact is.

Dr Maria Perez-Ortiz of UCL Computer Science, the report’s senior author, said: “AI systems are remarkably good at explaining scientific knowledge, but this can give us a false sense of their capabilities. Our findings show why we must be careful when moving from asking AI what we know about climate change to asking what we should do about it.”

View the original on Hindustan Times →

KioskNews shows a cleaned-up reading view extracted from the publisher’s page — the original always lives on their site, not ours.