AI in Healthcare and Public Services: Who Funds It, Who Benefits, and Who Is Accountable?

The AI Funding Files: Following the Money, Examining the Impact

Artificial intelligence is increasingly being considered for healthcare, local government, public administration and other essential services. It could help staff process information, reduce repetitive paperwork, identify patterns in data and give professionals more time to support the people who depend on them.

But when AI is introduced into a hospital, council or public service, the consequences can reach far beyond efficiency. A system may influence how quickly someone receives help, how their information is handled, how a decision is made or whether a vulnerable person is properly understood.

The UK Government's Budget 2026 announced a £15 million AI Challenge Prize, with the first competition focused on NHS challenges and a potential award of up to £2.5 million for the most promising solution. The Budget also confirmed a £50 million Civil Service AI transformation fund and an independent review of NHS productivity and AI, due to report in early 2027.

The important question is not simply whether AI can make public services faster. It is whether it can make them better, fairer and more accessible, while keeping people safe and ensuring that a real person remains accountable when something goes wrong.

1. Why are public services turning to AI?

Healthcare and public services face considerable demands. Staff may spend time writing notes, handling correspondence, processing applications, arranging appointments, reviewing records and managing large amounts of information.

AI tools may help with some of these tasks. Depending on the system and its intended use, they can summarise documents, support searches across information, assist with administrative work, help analyse images or identify patterns that professionals may wish to investigate.

However, AI is not one single technology. A tool that drafts a routine letter presents different risks from a system that helps interpret a medical scan or contributes to a decision about a person's eligibility for support.

Public bodies should therefore assess each application on its own merits. They should identify the problem, examine whether AI is appropriate and compare it with simpler alternatives before deciding to adopt it.

2. What funding has the Government announced?

Budget 2026 announced several initiatives connected with AI and public services. They have different purposes and should not be added together as though they were one healthcare AI fund.

• £15 million AI Challenge Prize: a competition to encourage UK-based innovators and businesses to develop AI solutions for major frontline public-service challenges. The first competition will focus on the NHS, with a potential top award of £2.5 million.
• £50 million Civil Service AI transformation fund: funding to support AI deployment across the Civil Service, reduce unnecessary administrative work and improve outcomes.
• £35 million for GovAI: funding to give civil servants access to frontier AI tools and capabilities and support departments in developing in-house expertise, procuring solutions and working with innovative companies.
• £60 million AI Pathfinder Fund: funding over 2026–27 to 2029–30 for practical interventions addressing cross-government barriers and opportunities arising from the AI transition.
• £30 million for local AI partnerships: support for locally led approaches to AI adoption, starting with authorities in West Yorkshire, South Yorkshire, and York and North Yorkshire.

Separately, the Government has committed up to £10 billion over three years to technology, digital and data initiatives supporting its wider NHS transformation plans. This is a broader commitment, not a dedicated £10 billion fund exclusively for AI.

These announcements indicate the scale of the Government's ambition. They do not, by themselves, demonstrate how much money has been spent, which systems have succeeded or whether the intended improvements have been achieved.

Source: HM Treasury, Budget 2026; Government response on AI regulation in healthcare.

3. What could AI do for the NHS?

AI may be useful in several areas of healthcare, provided that systems are appropriately tested, introduced and monitored.

• Administrative support: helping staff draft correspondence, organise information and complete routine documentation.
• Clinical decision support: providing information or identifying patterns that a qualified healthcare professional can review.
• Medical imaging: assisting with the analysis of scans or other images in approved and appropriate settings.
• Patient communication: helping people find information or navigate services, with routes to human support when needed.
• Research: supporting the analysis of complex datasets and helping researchers investigate potential treatments.
• Service planning: helping organisations understand demand, identify bottlenecks and plan resources.

These are possible uses, not guarantees of better care. A system that performs well in a controlled test may behave differently in a busy hospital, with different patient populations, incomplete records or unfamiliar working practices.

The National Commission into the Regulation of AI in Healthcare has emphasised that AI should augment rather than replace healthcare professionals. The aim should be to improve outcomes and support clinical judgement, not to introduce technology simply because it is available.

Source: National Commission into the Regulation of AI in Healthcare.

4. Can AI make public services more efficient?

Outside healthcare, AI is being explored for planning, correspondence, information management, customer enquiries and other administrative tasks.

Budget 2026 identified several existing or planned government applications:

• Ask HMRC: the Government reported 6.3 million customer interactions handled during 2025–26.
• Justice Transcribe: an AI-enabled probation tool being expanded across probation officers. The Budget reported approximately 267,000 hours saved to date.
• PlanAssist: a planning tool intended to become available to all councils in England by the end of 2027, subject to further testing. The Government aims to halve routine planning decision times.

These examples illustrate different stages of adoption. Reported interactions and estimated time savings are measures of activity or productivity; they are not sufficient on their own to establish the quality of the service or the benefits to the public.

For example, faster processing is not necessarily a success if more people receive incorrect information, have to repeat their applications or struggle to reach a human adviser.

Public services should measure speed alongside accuracy, accessibility, satisfaction, error rates, complaints, equality impacts and the amount of staff time genuinely released for more complex work.

Source: Budget 2026.

5. Who pays, and who benefits?

Publicly funded AI programmes may involve government departments, NHS organisations, councils, universities, technology companies and external suppliers. The funding mechanism matters because it determines what the public receives in return.

A grant may help a company research a new product. A prize may reward a successful solution to a defined challenge. A procurement contract pays a supplier to deliver a service. A wider technology investment may support infrastructure, staff training or digital systems.

Each arrangement needs a clear account of the money committed, the conditions attached to it, the outcomes expected and the risks borne by the public sector.

There may be genuine public benefits: better services, reduced administrative burdens, improved research capability, more effective use of staff time and opportunities for UK-based businesses. But commercial success does not automatically translate into improved public services.

Where private companies develop publicly supported technology, public bodies should understand the licensing terms, ongoing costs, data arrangements and ability to change suppliers. They should also consider whether successful projects can be reused elsewhere instead of requiring every organisation to pay to solve the same problem independently.

6. Patient safety cannot be an afterthought

Healthcare decisions can have serious consequences. An inaccurate AI output might contribute to a missed diagnosis, an inappropriate recommendation, a delay in treatment or a misunderstanding of a patient's circumstances.

AI systems can also perform differently across settings and populations. A tool tested on one group of patients may not work equally well for everyone. Poor-quality data, incomplete records and changes in clinical practice can affect performance.

On 6 October 2026, the Government published its response accepting the recommendations of the National Commission into the Regulation of AI in Healthcare. The programme includes work to strengthen accountability, improve monitoring and reporting of real-world problems, clarify responsibilities and support patients' access to redress.

The Government's response recognises that a one-off assessment before a product enters use may not be enough. AI-enabled medical products can change over time, and their performance may depend on the people, data, processes and organisations around them.

Healthcare providers need appropriate evaluation before deployment, staff training, clear procedures for reporting incidents and ongoing checks that a system remains safe and effective in practice.

Source: Government response to the National Commission's recommendations, 6 October 2026.

7. Who is accountable when AI gets it wrong?

One of the most important questions is who is responsible when an AI-assisted decision causes harm.

Responsibility may involve the system's manufacturer, the organisation deploying it, the professionals using it and the relevant regulators. The precise position depends on the technology, the circumstances and the applicable legal duties.

It should never be acceptable for an organisation to avoid responsibility by saying that the computer made the decision.

Public bodies should establish in advance:

• Who approves the use of the system and who oversees it.
• Which decisions AI may support and which must remain with a qualified person.
• How staff can question or override an AI output.
• How errors, near misses and complaints are recorded and investigated.
• How affected people can request an explanation or challenge a decision.
• How the organisation will suspend or withdraw the system if risks become unacceptable.

Human oversight must be meaningful. A person who is expected to approve hundreds of automated recommendations without enough time, training or information is not necessarily providing effective supervision.

Accountability also requires adequate records. When AI influences a significant decision, organisations need enough information to understand what role the system played and to investigate what happened if something goes wrong.

8. Privacy, consent and sensitive information

Healthcare and public services often handle highly sensitive information. AI systems may process medical records, personal circumstances, correspondence, images, voice recordings or information about someone's support needs.

Before using such systems, organisations must assess the applicable data-protection requirements and ensure that personal information is handled lawfully, securely and for clearly defined purposes.

Important questions include:

• What personal information is collected and why?
• Is the information being sent to an external supplier or used to improve a commercial model?
• Who can access the information, and how long is it retained?
• Can the supplier use it for other purposes?
• What happens if the system is breached or the information is inaccurate?
• Can the organisation operate the service without exposing unnecessary personal information?

Organisations should not assume that entering information into an AI tool is harmless simply because the tool is convenient. Staff need clear guidance about which systems are approved and what information may be entered.

In sensitive settings, privacy should be considered during procurement and system design, not added as an afterthought once a tool is already in use.

9. What about disabled and neurodivergent people?

AI may create useful opportunities for disabled and neurodivergent people, including support with communication, accessible information, speech-to-text, text-to-speech, organisation and creative expression.

It could also help professionals prepare information in different formats or make complex material easier to navigate. But these benefits depend on how systems are designed and used.

A tool that assumes everyone communicates in the same way may misunderstand someone who uses alternative communication, needs more processing time or expresses themselves differently. An automated assessment may also miss important personal circumstances that are not captured in its data.

For people receiving support, a system should not become a substitute for listening to their wishes, respecting their choices or understanding their individual needs.

Inclusive AI should be developed and tested with people who may be affected by it. This includes people with different disabilities, communication styles, levels of digital access and support requirements.

Accessibility testing should ask not only whether a person can open or operate the system, but whether they can understand its outputs, challenge mistakes and obtain human assistance when needed.

10. AI and safeguarding in public services

AI may help staff organise information or identify patterns that deserve further attention. However, it must not be treated as a substitute for professional judgement, safeguarding procedures or direct engagement with the person concerned.

Safeguarding decisions can involve complex relationships, communication differences, trauma, power imbalances and information that is incomplete or contested. A system may fail to understand context, repeat bias in its training data or produce a confident but incorrect explanation.

There are particular risks if automated tools are used to rank people by risk, flag families, interpret behaviour or influence access to support without adequate transparency and review.

Public bodies should define what the system is allowed to do, who reviews its outputs and how concerns are escalated. People must have a meaningful route to question decisions that affect their safety, rights or access to services.

For organisations working with children, disabled people and adults at risk, AI policies should sit alongside established safeguarding arrangements, staff training, confidentiality rules and clear reporting procedures.

The principle is straightforward: AI may assist people responsible for safeguarding, but it must not become the authority that decides whose concerns deserve to be heard.

11. Can AI reduce costs without reducing care?

One motivation for public-sector AI is productivity. If a system reduces repetitive tasks, staff may be able to spend more time on patient care, complex casework or direct support.

However, the financial case must include the full cost of implementation. Organisations may need to pay for licences, integration, computing, data preparation, cybersecurity, staff training, independent evaluation, maintenance and ongoing monitoring.

Some systems may also require extra human review because their outputs are unreliable or unsuitable for particular tasks. A tool that appears inexpensive at first may become costly if it generates errors, creates additional work or is difficult to integrate with existing services.

Any claim of savings should explain the baseline used for comparison, the period measured and whether the savings are cashable or represent time that staff can redirect to other work.

Most importantly, financial savings should not be treated as the only measure of success. Public services exist to meet people's needs. A cheaper service that becomes harder to access or produces poorer outcomes may represent a false economy.

12. What should a responsible public-sector AI programme include?

A responsible programme should be designed around the public need it is meant to address, with safeguards and measurable outcomes built in from the beginning.

• A clearly defined purpose: explain the problem the system is intended to solve and why AI is appropriate.
• Evidence before deployment: test performance in relevant settings and examine errors, limitations and equality impacts.
• Meaningful human oversight: make sure qualified staff have the time, authority and information needed to question outputs.
• Public and service-user involvement: include patients, disabled people, staff and other affected groups in design and evaluation.
• Data protection and security: assess privacy, access controls, data retention, supplier arrangements and incident response.
• Clear accountability: assign responsibility for approvals, monitoring, complaints and corrective action.
• Ongoing monitoring: track performance after deployment and review the system when data, workflows or circumstances change.
• Independent scrutiny: provide appropriate external evaluation, audit and public reporting.
• An exit plan: establish how the organisation can suspend the system, switch suppliers or return to a safe alternative.

13. How should success be measured?

Public bodies should publish meaningful measures of whether AI is delivering the outcomes promised. The appropriate indicators will vary by service, but could include:

• Waiting times and the time taken to complete administrative work.
• Accuracy, error rates and the frequency of human corrections.
• Patient or service-user outcomes and experiences.
• Accessibility and differences in performance across relevant groups.
• Staff workload, training and the time available for direct support.
• Complaints, incidents, near misses and how quickly they are resolved.
• Total cost of ownership compared with a credible alternative.
• Privacy, cybersecurity and compliance with relevant requirements.
• Evidence that the service remains effective after the system is updated.

Organisations should also publish limitations and failures, not only positive results. Transparent reporting helps decision-makers learn from mistakes and prevents a pilot that worked under limited conditions from being presented as proof that a system is ready for every setting.

14. Questions the public should be asking

• What specific problem is the AI system intended to solve?
• How much public money has been committed, spent and recovered in measurable benefits?
• Who selected the supplier and how was the decision assessed?
• What evidence shows that the system is safe and effective for the people who will use it?
• Who is responsible if an AI-assisted decision causes harm?
• Can a patient or service user challenge an outcome and speak to a real person?
• How are disabled people and other groups who may face barriers involved in testing?
• What happens to personal information processed by the system?
• Will the public be able to see independent evaluations and incident reports?
• What is the plan if the system fails, becomes unsafe or no longer offers value for money?

Conclusion: public services must remain accountable to people

AI may help the NHS and other public services manage information, reduce repetitive tasks, support research and improve access to some services. The Government's recent funding announcements and regulatory work show that AI adoption is becoming a significant part of its wider public-service agenda.

But public confidence cannot be secured through funding announcements or claims of efficiency alone. It requires evidence that systems work, that risks are managed, that information is protected and that people can challenge decisions affecting their lives.

For healthcare, the priority must be safe and effective care. For councils and other public services, it must be fair access, reliable decisions and meaningful support. Across all settings, human accountability must remain clear.

AI should help public services serve people better — not make it harder for people to understand, challenge or obtain help with decisions made about them.

The AI Funding Files examines public commitments, published evidence and the questions that should be asked as AI policy develops. Announced funding is not the same as money spent, and intended benefits are not proof of achieved outcomes.