The UK government is launching the Open-Source AI Fellowship, a 12-month initiative that will embed 10 top AI engineers and developers in public sector departments to co-develop practical tools using open-source models. Backed by a $1 million grant from Meta and administered by the Alan Turing Institute, the programme will begin in January 2026.
The Fellowship marks a shift from academic AI research to real-world deployment. Participants will work on projects such as enhancing national security language translation, speeding up housing approvals, and improving public-facing services including “Humphrey,” an AI summarisation assistant, and “Caddy,” a chatbot used by Citizens Advice.
“This is AI in action – open, practical, and built for public good,” said Peter Kyle, Technology Secretary. He emphasised the government's intent to move beyond theory towards deployable tools that improve efficiency and responsiveness. The programme also aims to strengthen the UK’s AI sovereignty by reducing reliance on proprietary US technologies.
Meta’s involvement goes beyond funding. By providing access to its Llama 3.5 model, the company is expanding its collaboration with UK public institutions. Meta stated that only publicly available, adult-only data is used, with users given opt-out options. The company underscored its compliance with UK data protection standards amid ongoing scrutiny of big tech’s role in public infrastructure.
The Fellowship supports broader government efforts, including the forthcoming AI Bill and the work of the AI Safety Institute, which independently assesses AI model safety. However, the participation of firms like Meta and Google continues to raise concerns about corporate influence over national digital policy. Kyle has acknowledged the need for a new kind of diplomacy to manage these complex relationships.
The programme has three main objectives: to foster in-house AI expertise, support talent development through direct government placements, and promote transparency via open-source tools. “Humphrey” and “Caddy” will be expanded and refined, reflecting the government’s preference for practical, user-focused AI applications. Applications are expected to open in late 2025, with fellows selected by Meta, the Alan Turing Institute and government representatives. The initiative runs in parallel with a separate Google Cloud partnership aiming to train 100,000 civil servants in cloud and AI technologies by 2030.
While the initial funding is modest, the Fellowship is a strategic step towards embedding AI capabilities in government. If successful, it could serve as a model for how to modernise public services with open and trustworthy technology.
Created by Amplify: AI-augmented, human-curated content.
