Is it safe to put client data into ChatGPT? What every firm should know
Putting client data into ChatGPT is safe only when three things line up: a business account that does not train on your data, data with client identifiers taken out, and controls that stop mistakes. Miss any one and the honest answer is no. A free or personal account with real client names, bank details or case papers is the risky case. A business plan with stand-ins replacing names is a different matter.
AI is welcome in a client-facing firm. Client data just needs to stay out of it. Here is how to tell the difference, and how to answer a client who asks.
Is it safe to put client data into ChatGPT? It depends on three things
The account. The data. The controls. Each one changes the answer.
The account
Not all ChatGPT plans treat your data the same way. According to Strac, on ChatGPT Free and Plus, OpenAI may use conversations to improve its models by default unless the user opts out in data controls. On ChatGPT Team, Enterprise and the API, business data is not used for training by default.
CloudEagle notes that OpenAI's enterprise privacy page says it does not train on data from ChatGPT Business, Enterprise, Edu or the API by default.
Deletion matters too. Usercentrics reports that under OpenAI's Enterprise terms, business data in deleted conversations is removed within 30 days, unless retention is legally required. OpenAI's data processing agreement also lets it keep processing de-identified or aggregated versions once individuals can no longer be identified.
The catch is simple. Staff often use personal accounts. A firm can pay for a business plan and still have Tom pasting into his own free login. Our guide on ChatGPT Team vs Enterprise vs free goes through each plan.
The data
A business plan reduces risk. It does not make every kind of data fine to share. Some data carries duties that no plan setting removes.
The controls
A written policy helps. People still forget it on a busy afternoon. Controls are what catch the paste before it is sent.
Rule of thumb: if ChatGPT does not need to know who the client is to answer the question, take the client out first.
Which client data is high risk
Some data is low risk. A general question about a tax rule. A request to tidy a paragraph with no names in it. Other data needs far more care.
| Type of data | Examples | Why it is high risk |
|---|---|---|
| Identity | Names with addresses, National Insurance numbers, passport details | Directly identifies a person |
| Financial | Card numbers, sort codes and account numbers, salary details | Can be misused and is hard to undo |
| Health | NHS numbers, medical notes, sick records | Special category data under UK GDPR |
| Legal correspondence | Client letters, case details, advice | Confidentiality and privilege can be lost |
| Lists of people | Client exports, CV batches, payroll sheets | Many people affected at once |
| Secrets | Passwords, secret keys, login details | Gives access to other systems |
What UK firms actually answer to
Most guides on this question stop at OpenAI's own terms. UK firms have their own rules too.
Solicitors and the SRA
The SRA (via Law Gazette) published a Warning Notice titled 'Misuse of AI' on 17 August 2026. It says using AI does not reduce a solicitor's professional responsibility. It flags two concerns: false or inaccurate AI-generated material, and confidential client information being entered into public AI tools.
The SRA has also warned that entering case details into tools such as ChatGPT will likely breach client confidentiality. Legal professional privilege may then be permanently waived and unable to be recovered. The SRA received 42 reports of AI-related wrongdoing between July 2025 and July 2026, covering issues including inaccurate citations, supervision and confidentiality.
Privilege, once waived, may not come back.
The courts have said similar. In UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), reported by Lexology, the Upper Tribunal observed that entering client letters and Home Office decision letters into an open AI tool such as ChatGPT may breach client confidentiality and could lead to privilege being waived. More in our guide on whether solicitors can use ChatGPT.
Every firm and UK GDPR
Accountants, advisers, recruiters and agencies all handle personal data. According to AI Policy Tracker (ICO AI guidance summary), UK GDPR Article 35 requires a Data Protection Impact Assessment before processing starts where AI processing of personal data is likely to be high risk. The ICO treats most AI involving profiling, large-scale processing or novel technology as meeting that bar.
A Data Protection Impact Assessment is a written check of the risks and how you reduce them. If you use a business plan, you also want OpenAI's processor terms in place.
How to keep the benefits and take the client out
ChatGPT rarely needs the real name to do the work. Drafting, summarising and rewording all work with stand-ins.
- Use a business account. Not personal logins.
- Swap names for labels. "Client A", "Supplier B".
- Take out numbers. Bank details, National Insurance numbers, NHS numbers, card numbers.
- Cut what is not needed. Paste the paragraph, not the whole file.
- Keep privileged papers out. Case details and advice letters stay in your own systems.
- Put the real names back yourself. After the answer comes back.
Worked example: Priya at Ashcombe wants a polite chasing letter. Instead of "Fenwick Joinery, invoice 4471, account 12345678", she writes "Client A, an overdue invoice". The question still goes. The letter comes back. She adds the real details in Word.
Doing this by hand works, but it relies on memory. Our guide to redacting data before pasting into ChatGPT covers the manual method.
Where a tool helps
Helio Guard does the swapping automatically on Windows PCs, in Chrome and Edge. It checks each paste, message and file upload on listed AI sites. Sensitive items are taken out and replaced with labels. Your own client names become stand-ins such as "Client A", and turn back into real names when the answer is copied. Checked on the PC, with fixed rules. We never see what your team types. It costs £1 a person a month, from 5 people, with a 14-day free trial and no card needed. Details are on our pricing page.
Common questions about client data and ChatGPT
Is ChatGPT Plus safe for client data?
ChatGPT Plus is a personal plan. On ChatGPT Free and Plus, OpenAI may use conversations to improve its models by default unless the user opts out in data controls. For client work, a business plan such as Business or Enterprise is the safer choice, and client identifiers should still be taken out before pasting.
Does a business ChatGPT plan make client data safe?
A business ChatGPT plan removes training on your data by default, which helps a great deal. It does not remove confidentiality duties, privilege risk or the UK GDPR need for a Data Protection Impact Assessment where processing is high risk. Firms still need to keep identifiers and privileged material out of prompts.
What should I tell a client who asks if we put their data into ChatGPT?
Tell the client plainly which AI tools the firm approves, which account type staff use, and what data is never entered. Explain that names and numbers are replaced with stand-ins before any question goes in. Our guide on replying to a client asking about AI controls has a ready structure.
Related: AI data protection for MSPs: how to package it, deploy it and sell it to your clients, Staff AI training: what to cover so people stop pasting client data, Is it safe to paste payroll data into ChatGPT? What to strip out first