Making AI operational in constrained public sector environments

AI Summary

A sponsored content piece published by MIT Technology Review's Insights division on April 16, 2026, examines how public sector organizations can operationalize AI through small language models (SLMs) rather than large language models (LLMs). According to a Capgemini study cited in the article, 79% of public sector executives globally express concern about AI data security, while a separate Elastic survey found that 65% of public sector leaders struggle to use data continuously in real time and at scale. Han Xiao, Vice President of AI at Elastic, is quoted extensively throughout the piece, highlighting key barriers including limited GPU infrastructure, restricted network connectivity, and strict data governance requirements that make cloud-dependent LLM deployments impractical or impossible for many government agencies. SLMs, which use billions rather than hundreds of billions of parameters, are presented as a viable alternative because they can be deployed locally, require less computational power, support auditability and regulatory compliance frameworks such as GDPR, and reduce reliance on centralized cloud infrastructure. The article references a Gartner prediction that by 2027, small, specialized AI models will be used three times more than LLMs, and cites an empirical study claiming SLMs performed as well as or better than LLMs in relevant tasks. Elastic's recommended approach emphasizes deploying AI-enhanced search capabilities over chatbot implementations, using techniques such as vector search, smart retrieval, and verifiable source grounding to process unstructured government data across mixed media formats.