1. Why AI is entering health and social care

Health and care services face significant demands on staff time, resources and service capacity. AI tools are being considered for tasks such as drafting notes, summarising meetings, organising information, supporting administrative work and assisting certain clinical processes.

In adult social care, the Department of Health and Social Care's September 2026 guidance recognises that AI may reduce administrative workloads and help staff spend more time with people receiving support. It also makes clear that AI will not always be the appropriate solution.

The important distinction is between using technology to support care and using it as a substitute for care. Saving staff time is valuable only if the time and resources saved contribute to better outcomes, safer services or more meaningful support.

2. Where AI could make a practical difference

• Administrative work: preparing draft meeting notes, action lists, routine documents and summaries for staff to check.
• Clinical support: assisting appropriately validated systems used by qualified professionals, where evidence demonstrates that they are safe and suitable for the intended purpose.
• Communication and accessibility: supporting some people with information presented in clearer language, alternative formats or other accessible ways.
• Planning and coordination: helping teams organise information and identify tasks requiring attention, without allowing an automated output to become an unquestioned decision.
• Staff learning: supporting training and access to relevant guidance, with appropriate checks on accuracy.

These are potential uses, not guarantees of success. Every tool needs to be assessed for its intended purpose, reliability, accessibility and effect on the people using or receiving the service.

3. Healthcare is not just a data problem

Healthcare decisions can involve complex symptoms, individual circumstances, uncertainty and competing risks. An AI system may identify a pattern or generate a suggestion, but its output may be incomplete, inaccurate or unsuitable for a particular person.

The National Commission into the Regulation of AI in Healthcare has emphasised that AI should augment rather than replace healthcare professionals. Its 2026 recommendations call for a more proportionate, risk-based approach that considers safety throughout a product's life, rather than relying only on a single assessment before deployment.

That matters because a system's performance can change when it is used with different populations, workflows, data or care environments. Monitoring, incident reporting and the ability to investigate problems are essential parts of responsible implementation.

AI should not be treated as a substitute for professional accountability, clinical judgement or a person's right to ask questions about their care.

4. Social care must protect dignity, choice and relationships

Social care is deeply personal. Support may involve communication, daily living, independence, relationships, emotional wellbeing and participation in the community. These needs cannot always be captured in a form, a digital record or an automated summary.

For disabled and neurodivergent people, technology can be useful when it reflects individual communication preferences, interests and goals. It can also create barriers when systems are difficult to use, fail to account for different needs or are introduced without involving the people expected to use them.

Digital tools should support person-centred care: the person remains central to decisions about their support. A system should not make assumptions about someone's abilities, behaviour, needs or preferences based only on a label or an incomplete record.

Services should offer meaningful ways for people to express preferences, understand what technology is doing and request human assistance. Digital options should not become a reason to deny an accessible non-digital alternative where one is needed.

5. The privacy question: what happens to sensitive information?

Health and care information can include diagnoses, medication, support plans, family circumstances, disability-related information and details of a person's daily life. Some of this information is special category personal data under UK data protection law and requires particular protection.

Putting confidential information into a public or unapproved AI tool can create serious risks. Organisations should not assume that a tool is safe for sensitive information simply because it is convenient or widely available.

Before adopting AI, providers should establish:

• What information the tool collects, stores and generates.
• Whether the provider is authorised to process that information for the proposed purpose.
• Who can access the information and whether it is shared with suppliers or used for further model development.
• How long records are retained and how they can be corrected or deleted when appropriate.
• Whether a data protection impact assessment is required.
• How security incidents, errors and inappropriate access will be handled.

Health and care organisations should involve the appropriate information governance lead, Data Protection Officer and Caldicott Guardian where relevant. Staff should follow their organisation's approved policies and must not enter identifiable or confidential service-user information into unapproved tools.

6. Human oversight must be meaningful

Human review is not simply asking a member of staff to click an approval button. A reviewer needs the time, knowledge, authority and information necessary to identify errors and challenge the system's output.

NHS England's information governance guidance highlights the importance of human involvement in significant decisions. Where UK GDPR protections for solely automated decisions apply, human involvement must be meaningful rather than a token gesture.

Good oversight means staff can:

• Understand the purpose and limitations of the AI tool.
• Check the original information rather than relying only on a generated summary.
• Recognise when an output is uncertain, biased or inconsistent.
• Override or reject an output without inappropriate pressure.
• Escalate concerns and record incidents.
• Explain decisions to the person affected and direct them to an appropriate review or complaints process.

Responsibility should be clear between technology suppliers, service providers, managers and qualified professionals. Buying a product does not transfer an organisation's responsibilities to the software.

7. Who is funding AI in healthcare—and what should the public expect?

AI in health and care may be developed by private technology companies, research institutions, public bodies and partnerships between them. Public investment can support innovation, evaluation, infrastructure, training and the adoption of digital tools.

However, an announcement that funding has been awarded does not prove that a system has improved care. Nor does a projected saving automatically mean that money will be reinvested in staff, community support or better access for patients.

Where public funding supports AI, people should be able to ask:

• How much public money was committed, and how much has actually been spent?
• Who received the funding, and what were they contracted to deliver?
• Was the technology independently evaluated in the setting where it will be used?
• What evidence exists of safety, effectiveness, accessibility and value for money?
• Were patients, disabled people, carers and frontline staff involved in its design or evaluation?
• Who owns the resulting technology and intellectual property?
• How will benefits, costs, errors and unequal outcomes be monitored after deployment?

Responsible investment should measure outcomes that matter to people—not only the number of tools purchased, tasks automated or staff hours supposedly saved.

8. Safety standards and accountability

In the NHS and related health technology settings, clinical risk management standards DCB0129 and DCB0160 address safety responsibilities for manufacturers and organisations deploying and using health IT systems. Organisations should establish what standards and regulatory requirements apply to the specific product and use case.

The National Commission's September 2026 report recommended a more lifecycle-based and system-wide approach to regulating and assuring AI in healthcare. On 6 October 2026, the Government published its response, accepting all 44 recommendations and setting out next steps. These recommendations and planned actions should not be mistaken for proof that every proposed change is already fully implemented.

Accountability also requires clear routes for raising concerns. Patients and people receiving support should be able to ask whether AI was involved in a decision, seek an explanation where appropriate and access the relevant review or complaints process.

9. An implementation checklist for health and care providers

1. Start with a real need. Define the problem and consider whether a non-AI solution would be safer, simpler or less costly.
2. Involve the people affected. Consult patients, people drawing on care, carers and frontline staff before selecting and testing a tool.
3. Assess risk and evidence. Check the tool's intended purpose, limitations, clinical safety requirements, accessibility and performance in the relevant setting.
4. Protect information. Complete the appropriate privacy and security assessments and agree clear arrangements with suppliers.
5. Train the workforce. Make sure staff know what the tool can and cannot do, and how to challenge its outputs.
6. Keep a human accountable. Define who reviews outputs, makes decisions and responds to concerns.
7. Monitor real-world outcomes. Track errors, complaints, unequal impacts, staff workload and whether promised benefits actually occur.
8. Be prepared to stop. Pause or withdraw a tool when risks outweigh benefits or its performance is not acceptable.

10. The AM-MediaSpace perspective: technology should serve people

At AM-MediaSpace, responsible AI means treating technology as an assistant—not an unquestioned authority. This principle matters in creative work, education, disability inclusion and safeguarding, as well as healthcare and social care.

People should not have to surrender privacy, choice or dignity to access a service that is supposed to support them. Disabled and neurodivergent people should be included in decisions about tools that affect them, rather than treated only as recipients of technology designed by others.

AI can be useful when it gives people more time, more accessible ways to communicate and better-supported choices. It becomes a problem when automation obscures responsibility, weakens relationships or makes decisions harder to challenge.

Conclusion: measure the improvement in care

AI may help healthcare and social care services work differently, but technological adoption is not the same as progress. The real test is whether it improves safety, access, independence, dignity and people's experience of support.

Public bodies, providers, developers and funders should be able to demonstrate what a system does, how risks are managed, who remains accountable and whether promised benefits have been achieved.

Healthcare and social care must remain centred on people. AI should support human care—not become an excuse to remove the human connection that makes care meaningful.

Sources and further reading

• Department of Health and Social Care — Using AI in adult social care (29 September 2026)
• National Commission into the Regulation of AI in Healthcare — recommendations for a future regulatory framework (10 September 2026)
• UK Government — National Commission into the Regulation of AI in Healthcare, including the Government response (6 October 2026)
• NHS England Digital — AI guidance for health and care professionals
• NHS England Digital — AI guidance for information governance professionals
• NHS England — Digital clinical safety assurance
• UK Government — A guide to good practice for digital and data-driven health technologies
• Department of Health and Social Care — Adult social care digital skills framework
• Data Protection Act 2018