CQC Won’t Inspect Your AI. It Will Inspect Your Care.

A registered manager rang me a few weeks ago with a question I have now heard in several forms: “Do we need to tell CQC we’re using AI?”

Her team had started using a tool to tidy up daily notes. Nothing dramatic. But she had read something about the regulator and AI, and she was worried she had wandered into territory without a map.

I understand the worry. I ran a number of care companies before Careberry, and I remember the feeling of a new rule arriving before anyone had explained it. So I read the Care Quality Commission’s statement on AI properly, start to finish, and I want to share what it says in plain English. Because the headline is simpler than most people expect.

CQC is not going to inspect your AI. It is going to inspect your care. Same as it always has.

What CQC published, and when

On 21 May 2026 the Care Quality Commission published its first formal position on artificial intelligence: “Artificial intelligence in health and social care: CQC’s role, expectations and plans”. It applies to England only. Scotland (Care Inspectorate), Wales (Care Inspectorate Wales) and Northern Ireland (RQIA) have their own regulators and their own approach.

Three things stand out.

First, CQC does not assess or approve specific technologies. It is not going to certify your rostering tool or your note summariser. There is no CQC badge for AI.

Second, CQC will judge AI through the regulations that already exist. The statement points to the Health and Social Care Act 2008 (Regulated Activities) Regulations 2014, and in particular Regulation 9 (person-centred care), Regulation 10 (dignity and respect), Regulation 11 (need for consent), Regulation 12 (safe care and treatment) and Regulation 17 (good governance). The duty under each of those sits with you, the registered provider. It does not transfer to a software company because the software is clever.

Third, on ratings, CQC says plainly that “the absence or presence of AI does not predict a specific rating.” You will not be marked up for using AI, and you will not be marked down for leaving it alone.

And a small detail that tells you where this is heading. Ring CQC today and the first thing you hear is a message saying your call may be recorded by an AI note taker. The regulator is using AI itself, and it tells you so before you have said a word. That is the standard it is asking of you: use it if it helps, and be open about it.

The 11 principles, without the jargon

CQC sets out 11 principles for providers using AI. I have grouped them the way I would explain them to a manager on a Tuesday afternoon.

Keep people in charge. AI should support decisions, not make them. Someone qualified checks what the AI produces, every time, and that checking is monitored rather than assumed.

Be open with the people you support. People should be told, in a way they understand, where AI is involved in their care, and offered a non-digital route if they want one. That is Regulation 9 and Regulation 11 in action.

Keep it safe and fair. The tool should give accurate results for everyone, not just the people it was tested on. If a summariser works well for one person’s notes and badly for another’s because of how they speak or what conditions they live with, that is a safety and equity problem.

Protect the data. Sensitive information must be stored and processed securely under UK GDPR and the Data Protection Act 2018. CQC specifically names a Data Protection Impact Assessment (DPIA) as a principle, not a nice-to-have.

Train your people. Staff should be confident using the tool and clear on where it fits in the care pathway, and where it does not.

Govern it like anything else that can cause harm. Risk assess it. Have a way to report, investigate and learn from failures. Know who is accountable if it goes wrong. Buy it in line with the standards that apply to you.

Read those again and notice something. None of them are new ideas. Replace the word “AI” with “a new medication system” or “a new agency” and every principle still holds. That is the point CQC is making.

What this looks like in a care home and in home care

In a care home, the most common use I see is note summarising and handover. A night team writes notes, an AI tool produces a summary for the day shift. The CQC question is not “did you use AI?” It is “did the day shift get an accurate picture of Mary, and could you show me how you know?” If a summary dropped a fall, or softened a refusal of food into “ate less than usual,” you have a Regulation 12 problem and a Regulation 17 problem. The AI is just where it happened.

In home care, the pressure point is different. Carers write notes on a phone, between visits, often in a hurry. AI can help turn rushed notes into something readable, and that is useful. But the person checking the summary may be an office coordinator who was not in the house. The human oversight CQC asks for has to be designed in, not hoped for. Who reviews it? When? What happens when the summary does not match the plan?

Either way, if an inspector asks “how do you know this is accurate?” the answer “the system does it” will not do. It never did.

Five things I would do this month

1. Write down where AI is used. One page. Which tools, what they do, who checks the output. If you cannot write it down, you do not yet have governance.
2. Complete a DPIA. CQC names it. The Information Commissioner’s Office (ICO) explains how. If your software supplier cannot help you with the data flows, that tells you something about the supplier.
3. Decide how you tell people. Add a plain sentence to your service user guide and care plan review. Offer the alternative.
4. Train the people who check the output. Not how to use the tool. How to spot when it is wrong.
5. Ask your supplier the hard questions. Where is the data processed? Is it used to train anyone’s model? Is the tool part of your assured care record, or a bolt-on nobody has assessed?

How we have approached it at Careberry

I will be direct about where I stand. I have written before that I use AI every day, and that most of what is sold to care providers as “AI” is hype. Both things are true.

We were early adopters of AI, both in how we build Careberry and in what we build for providers, and we did it with a rule: a person approves everything before it touches a care record, and every AI action is logged. Careberry AI runs inside the platform teams already use. AI Assist reads daily notes and flags risks, actions and errors, and reviews care plans for gaps. Knowledge AI answers staff questions from your own policies, citing the section. Capture AI, the one I am proudest of, turns an assessment or supervision conversation into populated care plan fields, with consent asked for and logged before any recording starts, and a coordinator editing and approving the draft. Nothing writes to a record on its own. And all of it is optional: AI is off until you switch a module on, and you can switch it off again whenever you like, service by service. That is the first of CQC’s 11 principles, “AI to support, not to replace,” built in before CQC wrote it down.

The data sits inside the same platform that is on the NHS Assured Solutions list, that meets the NHS Data Security and Protection Toolkit, and that is certified to ISO/IEC 27001:2022. When a provider asks us for help with their DPIA, we can answer the questions, because we built the system as a care provider who had to answer them ourselves.

People first. Process next. Technology always.

A closing thought

The manager who rang me did not need a new policy framework. She needed to hear that the standards she already works to still apply, and that her judgement still matters more than the tool. CQC’s statement says exactly that.

AI will keep improving. Regulation will catch up in pieces. But the thing an inspector will always want to see is the thing good managers have always been able to show: that you know the people you support, that your records tell the truth about them, and that a human is accountable for both.

If you would like to see how Careberry handles AI Assist, Knowledge AI and Capture AI, with the audit trail behind them, book a viewing of Careberry and we will show you, warts and all.