For UK charities · Case notes & client records
Not with real records about real, often at-risk people.
But you don't have to keep reading every note by hand. The fix is to summarise the same records somewhere private, where the summary is faithful to what's actually written and stays as accountable as the worker who relied on it.
Here's why case notes are the most sensitive data your charity holds, and how frontline teams get a trustworthy summary on private AI that never lets the record leave.
It usually starts as something reasonable. A caseworker is off sick and a colleague has to pick up their client, or there's a case review in the morning and months of notes to read first. So someone pastes the record into a public AI tool like ChatGPT, Gemini or Claude and asks for a summary. Nothing about the intent is reckless. The problem is what's in those notes, and where they've just gone.
Case notes carry health, mental health, abuse, addiction and immigration status, often details about children, all tied to a named and frequently at-risk person. The moment they're pasted in, all of it lands on a US company's servers and sits in logs you cannot see or audit. This is special category data at its most acute.
A public chatbot can quietly drop a risk, soften a warning, or invent a detail that was never in the notes. With a case record that isn't a typo. It can mean a colleague picks up a client without knowing the one thing that mattered most, and the worker who relied on the summary still carries the duty of care.
Case notes surface in safeguarding reviews, court, inquests, inspections and subject access requests. "A support worker pasted it into ChatGPT" is not an answer you want to give a serious case review, and there's no lawful basis for having sent the record there.
If a client asked under a subject access request what had happened to their record, or a reviewer asked how a summary was produced, would your charity have a clean answer?
None of this means reading every note by hand again. The summary your team needs is genuinely useful, and on a busy caseload it saves real time. The job is to produce it without handing the record to a stranger, and in a way you can trust.
Imagine pointing an assistant at the same client's notes, in plain English, and getting back a clear, structured summary, except nothing ever leaves your own systems and every line can be checked. Three things change.
The notes are summarised inside your own tenancy. No US company, no logs you cannot see, nothing handed to a third party to train on. Your clients' records stay exactly where they belong.
It's built to pull out the things that matter, the current situation, the risks and the outstanding actions, rather than smooth them away. It draws only from what's actually in the notes, so it doesn't invent a tidier story than the one on the record.
Every point in the summary links back to the note it came from, so a worker can check it in seconds and stays the one who decides. The assistant does the reading; the professional keeps the judgement.
That's private AI for case notes: the same summary your team needs, faithful to the record, on data that never leaves your control.
The build is a private AI assistant, a chat tool that looks and works like ChatGPT but runs inside your own systems, that reads a client's case notes where they already live, not something you paste records into. You ask for a summary in plain English; it reads the relevant notes and returns a structured picture, the history, the current situation, the risks and the recent actions, with each point linked back to the note it came from so you can verify every line. It reads only; it physically cannot edit, delete or export the record, and it draws only from what the notes actually say rather than filling gaps with invention.
Same summary you'd have pasted a client's record into ChatGPT to get. None of it leaving the building, and a summary you can stand behind.
I've built a sample assistant against a fictional client's case notes here, to demonstrate the kinds of thing a private AI build can do for you. The notes are entirely made up; nothing here is a real person. Below are a few everyday situations a frontline team might bring to it. Each one starts with what you're actually trying to do, then the question you'd type, then a short recording of the assistant answering and citing the notes it drew from.
A colleague is off sick and you have to pick up one of their clients today, with months of notes you've never read.
You'd open the assistant and ask:
"Summarise this client's case so I can pick up where my colleague left off."
There's a case review in the morning and you need the client's situation and current risks on a single page, not twenty entries to wade through.
You'd open the assistant and ask:
"Give me a one-page summary of this client's situation and current risks for tomorrow's case review."
You've been on leave for a few weeks and want to know what's actually changed for a client, rather than re-reading everything from the start.
You'd open the assistant and ask:
"What's changed for this client in the last month?"
Before you hand the client on you want to be sure nothing important is still hanging, and you want to see exactly where each point came from.
You'd open the assistant and ask:
"Pull out any risks or actions still outstanding from this client's notes."
Want to see it pointed at your own case management system, or have a question of your own in mind? Tell me what your team needs to get up to speed on →
You can paste text in, but you shouldn't with real client records. Case notes carry named individuals' most sensitive details, and the public version of ChatGPT sends them to OpenAI in the United States, keeps them in logs you can't see, and may use them to train future models. For a UK charity that's personal data leaving your control with no lawful basis to justify it.
No. Records about service users, especially involving health, mental health, abuse, children or immigration status, are personal and often special category data. Pasting them into a public tool sends them outside your charity entirely. You'd need a lawful basis, a contract with the provider and usually a data protection impact assessment; a paste into a public tool has none of those.
Yes, if the summarising happens inside your own systems rather than a public tool. A private build reads the record where it already lives and produces the summary there, so the confidential information is never sent to a third party in the first place. That's the difference between a tool that helps and one that creates a breach.
That's the real risk with a public chatbot, which can drop a detail or invent one. A private build is grounded in the actual notes and cites the source for every point, so the worker can check in seconds that nothing was softened or added. It draws only from what the notes say rather than producing a plausible-sounding story.
No, and it shouldn't. The assistant does the reading and produces a summary you can verify against the source notes; the worker stays the decision-maker and remains accountable. It's there to save the hours spent re-reading a record, not to make the call for you.
No. The build reads a copy of the record, so it can summarise but cannot write, delete or export anything. It also shows the note behind every line of the summary, so anything it tells you can be checked against the original.
Got a question that isn't here? Ask me directly →
Tell me who you are and what your organisation does. If any of this sounds like your situation, that's a good place to start. I'll let you know honestly whether I can help. Even a 30 to 45 minute call often leaves people with a clearer picture of the path forward, whether or not we end up working together. From there it's whatever fits: sometimes you don't need me, sometimes a short piece of scoping work makes sense first, and sometimes you already know what you want and we get straight to the build. There's no set process you have to follow.
For context: I work mainly with UK charities and non profits, with chief executives, operations and finance directors, programme leads, and the people who look after data and IT. Respectfully, I don't work with recruitment or development agencies.
Email: peter@peterbrady.co.uk
PNB Technologies Limited
55 Yew Tree Road, Ormskirk
Lancashire, L39 1NT, United Kingdom
Company Reg: 07166600
VAT Reg: GB986726850
© 2026 Peter Brady. All rights reserved.