The much-anticipated AI Opportunities Action Plan was announced today, with the Prime Minister warmly welcoming Matt Clifford’s recommendations for AI to deliver a decade of renewal. The plan sets out how AI can transform the lives of working people by, for example, speeding up planning applications, driving down admin for teachers (so they have more time to teach children), and spotting potholes to improve our roads. The plan is ambitious and backs AI ‘to the hilt’. Here’s our initial response to the plan, including where we think it hits the mark and where there are gaps.
Alignment with the AI Opportunities Action Plan
The ODI provided input at the consultation stage last summer, so we are pleased to see alignment with our work. Our vision for a National Data Library broadly aligns with the plan, where we called for using high-value public data assets and data stewardship models like federated Trusted Research Environments (TRE), including the NHS Digital’s TRE and the INSIGHT Health Data Research Hub. We also recommended open, interoperable data standards and strong data governance structures to ensure the quality of open government datasets.
The plan also sets out the intention to explore the use of synthetic data to ‘construct privacy-preserving versions of highly sensitive data sets’. We advocated for the use of privacy-enhancing technologies, such as federated learning and Solid, to preserve privacy when sharing data. Together with robust governance, these methods can ensure the transparency and accountability vital to gaining public trust, a cornerstone of both the AI Opportunities Action Plan and our vision.
What’s missing from the AI Opportunities Action Plan?
There are gaps in the plan. While it focuses heavily on technical infrastructure, it does not emphasise the socio-technical integration we advocate for. Specifically, the plan does not address the need for governance structures that combine technological safeguards with public oversight and user-centric approaches. It lacks detail on how user input will shape the design and implementation of the National Data Library, which is essential to ensure that diverse voices, especially those affected by AI or AI-enabled public services, are heard.
Furthermore, there is a lack of detail in the plan’s support for innovation, and it does not explicitly address the need for investment in data-centric AI, including specialist startups. While the action plan mentions developing guidelines for open datasets, it does not elaborate on establishing formal quality assurance or provenance tracking standards. We believe open standards and provenance documentation are essential to ensure data quality and usability.
The plan lacks detail on funding. There are several references to the need for investment by the government, but it's unclear exactly how these requirements will be met. So, we’ll look for further information on this in the Spring Spending Review, if not before. And finally, there is no mention of the Data (Use and Access) Bill currently passing through parliament, surely a legislative cornerstone for delivering so much of the plan.
To address many of these issues, we’re advocating for a ten-year National Data Infrastructure Roadmap to support the development of interoperable systems, AI-ready datasets, and privacy-enhancing technologies. This would support the plan’s focus on driving AI innovation through investing in long-term data infrastructure. The emphasis on high-quality data and strong governance in the AI Opportunities Action Plan closely aligns with our longstanding commitment to socio-technical solutions that integrate advanced data infrastructure with public trust.
The government as a data provider for AI
The government holds some of the world’s most valuable datasets. NHS health data for instance, has lifetimes’ worth of data for millions of people alone. Datasets like these across different sectors have driven scientific breakthroughs, powered innovation, and improved public services. But if they are to continue to do so, critical shortcomings must be addressed.
We recently published a report exploring the role of the government as a provider of data for AI. We identified that while AI tools such as ChatGPT and Gemini are increasingly relied upon for information on public services, they often fail to deliver accurate results despite scraping government data sources. Many tools use secondary or unreliable sources, such as social media, or make the answers up. This is because government data is often not published in formats that make it AI-ready. To drive the innovation in AI that the government desires, it must improve how it publishes its data.
Will the AI Opportunities Action Plan deliver?
The adoption of FAIR principles—making data findable, accessible, interoperable, and reusable—has long been championed by data.gov.uk and remains a strong foundation. Emerging tools such as Croissant, a machine-readable metadata format designed for machine learning, can help enhance discoverability and integration into developers’ workflows. Privacy-enhancing technologies (including Solid) can enable safe access to sensitive data while maintaining personal privacy, commercial sensitivity, and national security.
However, we hope the government will address some of the gaps in Matt Clifford’s recommendations as time passes. In particular, and to summarise, a National Data Library must be built with input and engagement from diverse stakeholders. Formal standards for data quality and provenance are critical for ensuring AI-ready datasets. Data-centric startups need explicit support to ensure robust data preparation and governance tools are developed. Finally, the importance of data at every stage of delivering the AI Opportunities Action Plan should be underpinned by creating, publishing, and delivering a National Data Infrastructure Roadmap. Only by improving data publication practices and investing in long-term data infrastructure can the UK position itself as a global leader in data provision for AI.
