Corporate report

UKHSA Advisory Board: Update on artifical intelligence in UKHSA

Updated 11 March 2025

Date: March 2025

Sponsor: Steven Riley

1. Purpose of the paper

The purpose of the paper is to update the Advisory Board of the ongoing activities supporting the adoption and utilisation of artificial intelligence (AI) in UKHSA and to seek input and guidance from Advisory Board members.

2. Recommendations

The Advisory Board is asked to:

• comment on the developments in our use of AI and approach to organisational readiness for adoption and utilisation of AI
• comment on UKHSA’s AI strategy

3. Activity since last update

Since the last update to Advisory Board, UKHSA has continued to explore opportunities to use AI to safeguard public health, both directly through innovative deployments of AI technology and indirectly by using AI to enhance business productivity allowing for resource to be redeployed elsewhere.

We continue to follow an approach to the adoption of AI through 3 pillars:

• enabling safe and responsible access to public AI tools
• evaluating enterprise level solutions for deployment at scale
• development of mission specific, public health use cases

We have expanded our portfolio of public health use cases delivered by technical specialists in Chief Data Officer (CDO) Group working closely with subject matter experts. Our recent update to the validation of large-language models (LLMs) for public health use cases highlights the success of this approach with a total of 16 classification and extraction tasks evaluated over three sub-domains of public health (burden, risk factors, and interventions).[1] These evaluations have focused on the performance of different LLMs for public health related tasks and have informed our selection of models on the internal high-performance compute cluster. Further data is required to quantify the benefits of these technical advances to the teams that are using them.

In addition, we have commenced an internal evaluation of i.AIs RedBox, a retrieval augmented generation (RAG) application that uses GenAI to chat with and summarise civil service documents. It is designed to handle a variety of administrative sources, such as letters, briefings, minutes, and speech transcripts.[2]  The application is being piloted on publicly available information whilst we seek approval to use sensitive documents.

At an enterprise level, we have completed an initial small-scale trial of M365 co-pilot in CDO Private Office and are developing plans to expand this trial in several other areas, including incident response. The results of the pilot are currently being evaluated with results expected by the end of March 2025. We continue to work closely with colleagues in Central Digital and Data Office (CDDO) and Government Digital Service (GDS) to respond to the outcome of the large-scale evaluation on M365 co-pilot. Similarly, several key suppliers are supporting the development of AI tools in their Software as a Solution (SaaS) applications and we are using these to develop the necessary approval and assessment processes through our Technology Front Door.

We are liaising across our technical teams to determine requirements for coding co-pilots. This will inform our strategic approach to providing capabilities to support efficient coding and best practice.

Notable successes of AI solutions that are in use or being piloted include:

Real-Time Pollen Monitoring

In collaboration with UK universities, UKHSA toxicologists have installed and are live-testing a cutting-edge commercial system integrating advanced data analytics with AI to detect airborne pollen in real-time. Combining sampling and analysis through automated imaging and AI technologies drastically reduces the time lag between sample collection and analysis from days or weeks to minutes. This allows for higher resolution, and more timely alerts, for key aeroallergen levels. Data will be gathered over the next 12 months. If tests prove successful, this technology has the potential to allow allergy sufferers and healthcare providers to better manage symptoms and/or prepare for high demands on health services.

Tuberculosis Screening Programme

The tuberculosis (TB) screening programme in the UK invites new migrants for testing based on country of birth. Country of birth is taken from GP registration records. However, this information is often ‘messy’ with incomplete data, misspellings, or ambiguous place names. This meant manual review of about 40,000 records was needed each year. Using AI can speed up this review by matching messy data to countries more quickly and in an automated manner. AI is shown to have 90% accuracy on deriving a country of birth and reduces the number of records that require full manual review to about 6,000 per year, representing an 85% reduction in manual effort. The system has now been integrated into the TB team’s regular workflow, and an Algorithmic Transparency Reporting Standard (ATRS) submission on the tool is being prepared.

The agency continues to believe that AI represents a significant opportunity to more effectively deliver its mission to safeguard public health.

4. AI strategy development

In August 2024, the AI Steering Committee (AISC), chaired by the Chief Data Officer, agreed that the agency-wide impacts of adopting AI necessitated the development of a dedicated AI strategy. The AI strategy sets out the strategic principles for adopting AI technologies over the next 18 months and describes the agency’s approach to creating an enabling environment to harness the benefits of AI to achieve our strategic goals. The AI strategy was discussed at ExCo in February 2025. Following discussion at Advisory Board, it will return to ExCo for final approval. The proposed AI strategy is provided as Annex One.

The AI strategy outlines 4 strategic principles which will guide our approach to maximising the benefits and managing the risks associated with AI:

• empower the agency to embed AI through reskilling our teams to ensure safe and effective implementation
• embed enterprise level AI use across UKHSA in alignment with recognised policies and risk management practices
• allocate funding to AI tools and products which tangibly improve business productivity, efficiency and data driven knowledge, policy and actions
• develop and learn from others’ uses of AI technologies that are aligned to our strategic priorities

The AI strategy has been aligned with the AI Opportunities Action Plan (AIOAP) to ensure that UKHSA is strategically positioned to take advantage of cross-government AI opportunities. A separate analysis exercise for the AIOAP has been completed to ensure the agency is aligned with the AIOAP, aware of progress to deliver the plan across government, and can influence and adopt relevant outputs.

The AI strategy will be accompanied by a delivery plan that sets out key outcome metrics and monitorable deliverables required to meet the ambition of the AI strategy and deliver on the strategic principles outlined above. To reduce additional governance burdens, deliverables will be owned by relevant business areas, and we will leverage existing structures and forums within UKHSA to monitor and assure delivery.

Work is ongoing to define and agree monitorable and assurable deliverables targeting the four strategic priorities. Examples under consideration include:

• working with heads of professions to understand learning requirements and deliver targeted learning
• implementing a Scan-Pilot-Scale approach to AI enterprise solutions
• establishing enterprise level principles for data management, unlocking potential for easier implementation of enterprise tooling
• implementing mechanisms to capture AI-related spend and investment across the agency

The success of the AI strategy will be evaluated through the AI Strategy Delivery Plan. We will identify approaches to assess the wider impact of AI transformation to demonstrate how the agency has changed because of its adoption of AI technologies. The impact and value of individual AI use cases will be evaluated in line with the principles set out in the AI annex to the Magenta Book.

As use of AI in the agency becomes part of our BAU processes, our approach to defining and documenting the agency’s AI strategy will change to reflect this. The next update of our Corporate Strategy will include detail on how the agency will use AI as part of our standard delivery strategy.

5. AI Readiness Agenda

The AI strategy is one of several products being delivered as part of the agency’s AI Readiness Agenda: an initiative overseen by AISC that aims to enhance business readiness for AI adoption.

The AI Readiness Agenda covers the full range of areas where consideration of AI’s impact on the agency is required, including:

• the impact on our people and culture
• our tooling and infrastructure needs
• how we can ensure secure and responsible adoption
• how we can assess and evaluate our AI use

A full list of the areas in scope can be found in Annex Two. Where possible we have adopted or adapted existing material. This includes adapting both established business-as-usual (BAU) processes, and relevant material from central government and industry. We are maintaining a flexible approach to allow the agency to respond to the wide range of guidance published by central government, for example the AI Playbook and AI annex to the Magenta Book.

We anticipate that AI readiness products will be delivered and handed over for embedding BAU functions in quarter one 2025 to 2026. Delivery of the AI readiness products will provide the agency with a strong foundation to build on with delivery of the AI strategy.

Given the complexity and potential impact, successful adoption of AI requires cross-agency collaboration. Work on the AI Readiness Agenda to date has brought together key stakeholders from across the agency at both SCS and delegated grades. We anticipate that this collaborative approach will continue as we move into adoption.

6. Further cross-agency activity and challenges

The agency’s recent restructuring activity, alongside the continuing requirement to deliver its BAU remit, has limited its capacity to absorb change. This has restricted the speed at which AI readiness and adoption can proceed.

Whilst Advisory Board has previously indicated a preference for the agency to increase its risk appetite for the use of AI internally, the agency’s ability to do so is limited by operational delivery factors. These include resource constraints, technical limitations, and requirements to ensure AI use is compliant with cyber, privacy, and ethical frameworks. We anticipate that continued delivery of the AI Readiness Agenda and AI Strategy will alleviate these limitations.

We continue to encounter challenges in deploying enterprise level solutions, due to the agency’s fragmented data structures, inconsistent labelling and classification of data, and lack of consistent permissions. Addressing these issues to enable enterprise level AI tooling is part of the AI Strategy Delivery Plan and wider information management activities.

Adoption of AI in the agency’s scientific activities is included within the Chief Scientific Officer Group Transformation Programme (CSO Transformation). We will continue to work with colleagues in CSO Group to ensure that adoption is in line with the AI strategy.

We are aware of further interest across the agency in harnessing AI to tackle specific business problems, produce innovative outputs, and enhance effectiveness of day-to-day delivery. Our AI Readiness Agenda has delivered the initial changes to the agency’s policies and process to enhance business readiness for AI adoption. This includes a communications approach to engage our people in discussion of AI adoption at workforce and profession level to develop an AI-positive culture. As part of the delivery of the AI Strategy, we will make further changes as necessary to ensure access to AI solutions is streamlined, secure, and supports the agency’s strategic objectives.

As previously reported to Advisory Board, an AI Register has been developed to capture information on AI projects developed in-house. The scope of the AI Register is currently limited to projects developed by the Analysis and Intelligence Assessment Directorate. Pilots are underway to inform the planned rollout of the register, with the ambition that all internal UKHSA AI projects will eventually be captured.

Health security risk assessment

We continue to monitor the potential impact of AI on public health security. The All Hazards Intelligence – National division within the Analysis and Intelligence Assessment directorate has refreshed its assessment of the key risks to health security posed by AI, and the likelihood of those risks materialising in the next 12 months. The assessment, provided as Annex Three, was prepared in conjunction with colleagues across government and will support our management of the identified risks.

7. Government Internal Audit Agency report and recommendations

The Government Internal Audit Agency (GIAA) carried out a risk-based audit of UKHSA’s approach to AI which assessed our governance framework, strategic approach to AI and risk management framework.  The field work was carried out in June and July 2024 and the audit report reflects a view of the agency’s position at that time.

The overall finding that UKHSA has ‘moderate’ controls in place to mitigate the above risks represents the second highest rating on GIAA’s four-point rating scale. GIAA commended the agency’s progress in establishing AI governance and risk management frameworks. They noted that the evolving nature of AI and limited guidance from central government made the frameworks immaturity understandable and that further work was required to sufficiently provide AI controls.

GIAA made eight recommendations to strengthen UKHSA’s controls and we remain on track to complete all related actions by the end of September 2025.

Nick Watkins, Deputy Director, Data Science and Geospatial; Chief Data Scientist

---

[1] Harris and others. Evaluating Large Language Models for Public Health Classification and Extraction Tasks

[2] i-dot-ai/redbox: Bringing Generative AI to the way the Civil Service works