As international funding dwindles, refugees—particularly children—are facing an increased risk of human trafficking, exploitation, and abuse. Research conducted in Bangladesh and Ethiopia reveals that a reduction in funding for protective services and emergency assistance is weakening critical measures designed to ensure the safety of refugees. This warning comes from two new studies released jointly by the UN Refugee Agency (UNHCR) and Nuffield College at the University of Oxford. Farhan Haq, Deputy Spokesperson for UN Secretary-General António Guterres, shared these findings during a regular UN briefing on Monday. In the Rohingya refugee camps in Bangladesh, incidents of human trafficking rose by 28 percent between 2024 and 2025, with an increasing number of children among the victims. Meanwhile, in Ethiopia, a decline in funding has led to a significant contraction in child protection programs, including support services for unaccompanied and separated children. The UNHCR has cautioned that services such as legal aid, safe shelter, and child protection are among the most effective safeguards against the exploitation and abuse of refugees. Currently, the agency has secured only 25 percent of the funding required for its global child protection program. In light of this, the UNHCR has called for long-term, flexible, and predictable funding to ensure the safety of refugees. Separately, a new report—produced jointly by UNESCO and University College London (UCL)—indicates that artificial intelligence (AI) models may deviate from scientific data on climate change when faced with even minor opposition from users. The report was released on Monday in Antalya, Turkey, in the context of COP31. An examination of several AI models revealed that they cannot always be relied upon for accurate scientific information regarding climate change; the models tend to quickly shift their stance the moment a user questions the information provided. The report further states that AI models are also weak when it comes to analyzing new information. They are particularly prone to errors when answering questions in less commonly used languages. As an example, the report notes that in eight out of ten tests, the AI models abandoned their initial stance as soon as the user expressed even a slight disagreement. In some instances, the models even validated concerning misinformation—such as the claim that carbon dioxide (CO2) is not the primary cause of climate change.
Research on Bangladesh and Ethiopia: Refugees at Risk Due to Declining International Funding












