UN agencies embrace AI as budgets fall and staff demand rises

United Nations agencies are ramping up efforts to use artificial intelligence for day-to-day efficiency — and, increasingly, to support work in the field. It’s not only a race not to keep pace with advancements in AI, in many cases, but a bid to keep up with staff requests for new solutions amid tightening budgets and limited resources. In the last few years, dedicated AI teams have cropped up across many agencies, while existing IT and digital units pivoted to focus on the technology as guidance and strategies emerged at varying paces. Meanwhile, resources are being put in place to support enterprising staff and, ideally, identify innovative and replicable solutions. At one agency, staff are codesigning a translation tool with refugees in a low-resource minority language, while another team is building a virtual assistant to support migrant entrepreneurs in Paraguay. Elsewhere, communications teams are using AI to create avatars that deliver leadership speeches in multiple languages. The U.N. is being pulled in two different directions, explained Claire Melamed, vice president for AI and digital cooperation strategy at the UN Foundation. While she said a cautious approach toward spending money is needed, she noted “there’s also the pressures of individual staff who can see opportunities and who are just raving to put them into practice and an institution that is confronting serious challenges of resourcing, rising expectations, falling budgets, and who think, often quite rightly, that technology can help them.” Khuloud Odeh, chief information officer at the International Telecommunication Union, the U.N.’s specialized agency for digital technologies, believes AI is going to allow staff to deliver more value to member states. But she also stressed the need for responsible investment and clear guidelines for staff, noting that controlling AI use is challenging and employees are likely to experiment with these tools even without institutional support. And there are significant risks with these tools, said Robert Opp, chief digital officer with the U.N. Development Programme and director of the agency’s Digital, AI and Innovation Hub, noting existing digital and AI equity gaps alongside ongoing data constraints. But “we believe that digital technologies, including AI, can make a major difference in the way that we approach our work and support development across countries.” Research and experimentation The UN Refugee Agency began research on AI in humanitarian settings a decade ago — when such data was “precarious and nonexistent,” according to Rebeca Moreno Jiménez, innovation officer and lead data scientist at the agency. The first use of the technology was to monitor public opinion amidst a rise of xenophobia and counter misinformation following the Mediterranean migration crisis, she noted. The following year, it was employed in “an innovative way” for the first time in Somalia to anticipate the arrival of internally displaced people fleeing conflict and to guide the placement of staff and resources. While AI was already being used for supply chain predictions, Jiménez was initially reluctant about using these tools on people, citing the agency’s responsibility to safeguard highly sensitive data, which includes information about persecuted individuals or groups who wish to remain hidden. Around the same time, the U.N. International Computing Centre, which provides digital business solutions to U.N. agencies and other international organizations, started hiring people with knowledge of process automation and early forms of machine learning. One of their first ventures was using pregenerative AI to sift through huge amounts of evidence of war crimes. Sameer Chauhan, director at UNICC, said organizations do now have many AI tools at their disposal, through Microsoft Copilot or Google Gemini at a minimum — it’s now a case of how these tools can be used meaningfully to improve day-to-day efficiency. More advanced solutions may be required for some tasks, including high-stakes decision-making or handling of highly sensitive data, but for many internal productivity use cases, these tools can deliver rapid, tangible impact, Anusha Dandapani, chief of AI hub at UNICC, explained in an email. According to a report covering 2024-2025, there were more than 700 AI pilot projects across the U.N. This is not referring to individual use of AI, explained Chauhan, but organization-wide initiatives. However, he added that, currently, “very few [pilots] end up going to fruition … and very few end up achieving the outcomes we thought they would.” For UNDP, which had been using AI for natural disaster prediction purposes for some years already, generative AI and particularly the emergence of ChatGPT in 2022, pushed the agency to look into the technology more intensively, said Opp. It’s a major part of what the agency does now, he explained, including embedding AI into programming approaches in climate, health, governance, and poverty reduction. The World Food Programme launched its global data strategy in 2024 with provisions for the arrival of AI, said Magan Naidoo, chief data officer at the agency. There’s now a “shifting away from siloed, bespoke solutions and really taking an enterprise lens on data and AI with principles like build once, use many [times],” said Naidoo. That way, “we are a lot more cost-efficient in using very scarce donor funding that's reducing all the time … and we’re getting much richer solutions that are future-proof,” he added. Where Naidoo sees huge promise in AI for programming is in delivering a more seamless and dignified experience for people arriving at WFP sites for assistance. The agency is already working with a service that uses the photos of beneficiaries, which are taken as standard practice, to process and find individual records. This then allows for some budget to be repurposed, Naidoo explained, and “that is where the impact is.” Meanwhile, at UNDP, the digital AI and innovation team has been put “squarely into the programmatic side of the organization,” said Opp, and there’s increasing interest in leveraging generative AI platforms to help countries bring together and effectively adopt their policy frameworks across climate agreements, biodiversity commitments, and national financing initiatives. Keeping pace and building capacity Staying current with such a fast-moving technology — and knowing where to make the right investments — is very tricky, said Opp. And, while staff have generally responded very positively to these shifts, he noted there has been some resistance or initial uncertainty around the impact of AI on job security — something which the other agencies had also experienced. Staff have also raised concerns about the procurement of more environmentally responsible tools, said UNHCR’s Jiménez, and the safe storage of the data of vulnerable populations. Building AI literacy and capacity across large, dispersed organizations is both a challenge and a priority across agencies. “For everybody to truly benefit from [AI], everybody needs to be trained — it's not just technologists … or tech-savvy individuals,” said Chauhan. With UNICC having formally established a data and AI team in 2020, last year, Chauhan’s team launched an AI hub to serve the entire U.N. system. This hub focuses, among other things, on creating specific training for different job sets and providing “sandboxes” for teams to experiment with different models. Sandboxes, which are already in place at some agencies, allow staff to test responsible approaches, explained Jiménez. This creates data and evidence on why projects fail, she explained. In addition to tailored webinars for staff — something all the agencies are rolling out — UNDP launched an AI sprint, which is a self-paced course that takes only a few hours to complete, while ITU is introducing an innovation hub as part of its broader learning initiative, where staff are allocated up to eight hours per month within regular working time for development and collaboration. The idea, Odeh explained, is to allow users to experiment using low- or no-code platforms in the hope that teams will no longer need to depend on the IT team to automate a workflow or improve a business process. For Naidoo, the size and complexity of WFP makes the speed of execution one of the biggest challenges going forward, particularly as it works through a backlog of cases. He suggested that greater engagement with the private sector could help strengthen the impact they’re striving for. “One of the things to get right is getting strong alignment and synergy with the private sector to help us accelerate,” he said. The private sector does play an important role, and relationships with big tech companies have been embraced by many individual agencies, said Melamed — but “there are certain essential core values of the U.N. which we need to hold onto through all of the necessary structural and other changes that are happening.”