º£½ÇÉçÇø

As AI adoption accelerates across government, civil and public servants working on real-world applications are encountering a common set of challenges.

Over the past nine months, 23 senior civil servants participating in the ImperiaPolicy Forum’s AI Policy Fellowship have examined what it takes to turn AI’s potential into meaningful public sector impact.  

This year’s Fellows came from a diverse range of government departments and public-sector organisations, bringing expertise spanning technology, defence, public health, transportation, diplomacy, local government, Parliament and more. Throughout the Fellowship, they explored how AI can transform public services through research projects developed in close collaboration with members of º£½ÇÉçÇø’s academic and research community working at the cutting edge of AI and innovation.

During the Fellowship’s concluding session, Fellows came together to present their work and reflect on shared challenges and best practices. Their discussions highlighted the importance of connecting technical innovation with the realities of policymaking and public-sector delivery, set against the ongoing debate about the wider implications of AI for society. 

Several key themes emerged from these discussions, including: the resources and expertise required to effectively adopt AI; the importance of sound leadership and governance; the need to rethink established approaches to digital transformation; and the challenge of evaluating AI’s impact in complex public-sector environments. 

Together, these point to a broader shift in how the UK government is approaching AI. Civil servants are no longer simply considering the hypothetical question of what AI means for government, but how can government harness AI effectively and responsibly. 

Understanding AI as a system-wide challenge 

The session opened with a module on the risks of agentic AI led by , Associate Professor in Computer Science and Director of º£½ÇÉçÇø’s Academic Centre of Excellence in Cyber Security. Fellows examined the complex security risks of agentic systems—essential considerations for government. 

The session also included a fireside chat between George Margereson, Director of the new AI Taskforce in the Cabinet Office and a Fellowship alumnus, and , Professor of Pure Mathematics and academic lead of our AI Fundamentals course. The conversation explored the Taskforce’s work in responding to the breadth and complexity of the policy challenges associated with AI. 

The conversation highlighted how AI does not sit within a single policy area. Decisions about innovation and AI have implications for economic growth, labour markets, energy demand, digital infrastructure and public services. These issues cannot be addressed in isolation; one of the main objectives of the Taskforce is to develop a strategic approach for the UK. 

Fellows’ projects 

During the day, Fellows also shared insights from their research projects. Presentations from civil servants at the AI Security Institute, UK Health Security Agency, Ministry of Defence, Food Standards Agency and more covered the following critical topics: 

• Evaluating how AI agents perform in high-stakes settings 
• Understanding the impact of AI on roles within civil service professions 
• The importance of robust evaluation for AI pilots 
• How AI chatbots can support public-facing advisory work 

Imperiaacademics and researchers who mentored Fellows throughout their projects also had the chance to gain new perspectives on how AI research can directly inform and shape government decision-making. 

Three lessons on AI adoption across government 

Each of the Fellows’ projects brought the practical challenges of government AI adoption into focus. Three connected lessons emerged: the resources and expertise required to develop AI and use it effectively, the importance of evaluating AI in real-world contexts, and the need for strong leadership and governance to guide adoption. 

1. Developing and deploying AI requires significant resources and expertise

Developing even small-scale AI pilots can be highly resource intensive, often dependent on access to technical tools and infrastructure, as well as adequate time, funding and organisational capacity. In one participant’s experience, piloting an AI-enabled regulatory adviser required the extremely time intensive—but valuable —process of training models on a reference library built on authoritative and human-verifiable expertise. 

Training the model must also be done with and by people who understand the relevant policy, operational and service context. This expertise is essential for ensuring these tools are accurate and responsive to the realities of public-sector work. 

2. Evaluating AI means looking beyond the output

Technical performance alone cannot determine whether AI is worth adopting. For example, a tool may perform well in a controlled pilot but may still require significant resources to properly evaluate whether it can introduce new risks or fail to deliver meaningful improvements in practice.  

Different teams may also have different expectations of what AI ‘success’ looks like, from increased productivity to improved service quality. Without a shared understanding of these objectives, it can be difficult to compare results and identify next steps. In one participant’s experience evaluating AI pilots, the lack of standardised metrics for ‘success’ across teams risked a breakdown of aligned and effective evaluation practices.

3. AI adoption across government and public services requires strong leadership and governance

AI is also changing the game for digital transformation. While previous widespread technology delivery focused heavily on implementation, AI’s rapidly evolving capabilities raise wider, less predictable questions about accountability and risk. This makes leadership and governance central to successful adoption.  

Organisations need clear responsibility for deciding where AI should be used, what risks are acceptable, how systems should be monitored and when human intervention is required. They also need governance arrangements that can adapt as the technology evolves. Fellows broadly agreed AI adoption requires leaders to connect innovation with institutional priorities and capacity building. 

Looking Ahead: Building capability for the next phase of AI adoption 

As government departments and public-sector organisations continue to explore different AI applications, the priority will be to drive meaningful, responsible and evidence-led progress.  

The ImperiaPolicy Forum will continue to support this work by connecting policymakers with º£½ÇÉçÇø’s research community. We create opportunities for civil and public servants to explore practical AI questions and strengthen the links between technical innovation and public policy.  

Applications are now open for the 2027 AI Policy Fellowship. Join a growing network of leaders and innovators at the intersection of AI and public policy and help shape the future of AI in government.