AI in adult social care: what CQC expects, and where dignity comes first

Cue 1 Two families of AI in care: pain estimation and sensors on one side, drafted notes and rotas on the other

EMMA 00:00 George, our care home's new records system writes the notes for you. You say a few words and out comes a paragraph. Is that allowed?

GEORGE 00:10 It's allowed. Whether it's safe depends on what you do next. Let me set out where AI turns up in care first. There's guidance for providers from the Digital Care Hub, published in 2025 and commissioned by NHS England's Digitising Social Care programme.

EMMA 00:27 What does it cover?

GEORGE 00:28 Two families. In care delivery, tools that estimate pain from a person's face, and sensors that detect movement, sound or a fall. And in administration, software that drafts notes and plans, writes rotas, summarises documents.

Cue 2 The CQC statement dated 21 May 2026 beside Regulation 12 and Regulation 17, with no new rulebook

EMMA 00:42 And what does it recommend?

GEORGE 00:44 Good data going in, secure and ethical use, human oversight, and training. It even includes a template AI policy.

EMMA 00:52 What about CQC? Do they have rules on this?

GEORGE 00:56 They published their first statement on 21 May 2026. The central point is that there's no separate AI rulebook. They look at AI through the existing regulations. Safe care and treatment, Regulation 12. And good governance, Regulation 17.

Cue 3 Six expectations as six cards

EMMA 01:11 So using AI won't mark us down.

GEORGE 01:14 Not in itself. How it's governed, checked and explained is what counts. A service using a tool nobody understands is worse off than one using no tool.

EMMA 01:24 What do they actually expect?

GEORGE 01:26 Six things, in summary. AI supports human decisions and doesn't replace them. Trained staff check what it produces. People are told and given a choice, with an alternative that doesn't depend on it.

Cue 4 A drafted care note with ate well highlighted and a carer correcting it, with four things to check

EMMA 01:40 That's three.

GEORGE 01:40 It's safe, reliable and fair for everyone. It's secure and lawful with data. And it's governed. Staff trained, a data protection impact assessment done, and a clear way to report when something goes wrong. Read the statement itself for the exact wording.

EMMA 01:57 Back to my notes. What can go wrong with a drafted one?

GEORGE 02:02 It reads well and it may be wrong. It can add something usual that didn't happen. Ate well, when she left most of it. It can leave out something that did happen. Or change a detail.

Cue 5 A personal phone with an unapproved app crossed out, and a care worker telling their manager

EMMA 02:16 So what do I check?

GEORGE 02:18 Four things before you save. Medicines. Falls and injuries. Food and fluids. And what the person said or chose. Then ask what happened that isn't there.

EMMA 02:29 And if I'm on a home visit and use an app on my own phone?

GEORGE 02:35 Don't, unless your employer has approved it. A tool you chose isn't covered by their data protection arrangements, and care information is special category data. The strongest protection in law.

EMMA 02:47 I think some of the team already do.

GEORGE 02:50 Then they should tell the manager straight away, and the manager should make it easy to. A care record is evidence of what happened to a person. If you wouldn't have written that sentence yourself, don't save it.

Cue 1 A carer explaining a sensor to a resident in plain words, with five things to cover and a clear option to decline

EMMA 03:05 George, we've had acoustic sensors fitted in the bedrooms. They listen for falls at night. Nobody asked the residents. Is that a problem?

GEORGE 03:14 Yes. CQC expects people to be given information about the part AI plays in their care, and to have a choice. With an alternative that doesn't rely on the technology.

EMMA 03:26 What do we have to tell them?

GEORGE 03:29 In ordinary words. What it is, what it does, what it records, who sees it, and that they can say no. At a moment they can take it in. Not buried in an admission pack.

Cue 2 The Mental Capacity Act principles beside a sensor, with all night crossed out and two hours after medication chosen

EMMA 03:43 And if they say no?

GEORGE 03:45 Then it has to be a real no. If declining the sensor means a worse service, it wasn't a choice.

EMMA 03:53 Half our residents couldn't make that decision.

GEORGE 03:56 Then the Mental Capacity Act applies, as it does to any decision. Assume capacity. Support the person to decide. An unwise decision isn't a lack of capacity. And if they can't decide, act in their best interests and choose the least restrictive option.

Cue 3 A relative's request beside a best interests decision record

EMMA 04:13 What does least restrictive mean for a sensor?

GEORGE 04:16 It's a real question. Is it needed all night, or for the two hours after medication? Is a camera needed when a pressure mat would do? Monitoring restricts privacy, so less is better where less will do.

EMMA 04:31 One daughter asked us to put a camera in.

GEORGE 04:35 A family member asking isn't consent. Unless she holds a relevant lasting power of attorney or deputyship, it's a best interests decision the service must make and record. And continuous monitoring can be part of what makes someone's arrangements a deprivation of liberty, which needs proper authorisation.

Cue 4 The same sensor shown twice: sparing a person from being woken, and replacing night staff who never come

EMMA 04:53 Is all this monitoring a bad thing, then?

GEORGE 04:57 Not at all. There's a useful test. In 2024 care workers and researchers, brought together by Oxford, the Care Workers' Charity and the Digital Care Hub, defined responsible use as supporting and not undermining human rights, independence, choice and control, dignity, equality and wellbeing.

EMMA 05:14 Give me an example either way.

GEORGE 05:17 A sensor that means someone isn't woken every two hours by a torch at the door adds to their dignity. The same sensor, used to justify fewer staff at night with nobody coming when it sounds, takes it away.

Cue 5 Four reasons to stop and ask, and a route from manager to whistleblowing to the regulator

EMMA 05:32 We also have an app that reads pain from a face.

GEORGE 05:37 Then know that tools can be less accurate for some people than others. It may have been tested mostly on faces unlike the person in front of you. If what it says doesn't match what you see, trust what you see and report it.

EMMA 05:54 And when should I refuse to use something?

GEORGE 05:57 When it's used without the person knowing. When it stands in for a check a human should make. When its output doesn't match the person you know. Or when nobody's approved it.

EMMA 06:10 Who do I tell?

GEORGE 06:12 Your manager first. Then your whistleblowing route, and you can contact CQC directly. A concern about a tool is a concern about care. Ask of anything new: does this help me be with this person more, or does it stand in for me being there?