Is it safe to put client data into ChatGPT? A UK GDPR checklist

Putting identifiable client data into a personal ChatGPT account is not safe for an accounting firm. It risks breaching client confidentiality and UK GDPR, and ICAEW has warned members against it. A business plan with a proper contract, a data protection impact assessment and anonymised prompts can make some uses acceptable.

What does ICAEW say?

ICAEW’s position is that client data in a public AI tool is a confidentiality risk. Its article Don’t let AI tools land you in hot water, published on 12 February 2024, says that if information you put into AI tools includes client data, “it’s very likely that this will breach client confidentiality”.

The same article warns that AI tools “hallucinate” and invent references. It advises members to follow their own firm’s guidelines and to “retain professional scepticism”. Confidentiality applies to all client information, not only personal data, so a company’s figures count too.

How do ChatGPT plans differ?

The plan you use changes what happens to your data. OpenAI’s own help centre and business data pages state the following; we could not open them directly on 6 November 2026, so check the live terms before you rely on them.

On personal plans, the setting is under Settings, Data controls, “Improve the model for everyone”. Turning it off affects only future chats. OpenAI also says data residency, including in the UK, is available for Enterprise, Edu and API customers, not for personal plans.

What does UK GDPR require?

Your firm is the controller of the client personal data it holds, so you stay responsible when you use an AI tool. These rules apply.

A processor contract. Article 28 UK GDPR says you may only use processors giving “sufficient guarantees” and that the processing must be under a written contract. The contract must require the provider to act only on your instructions, keep data secure, and delete or return it at the end. A personal account accepted under consumer terms is unlikely to meet this.

A DPIA where the risk is high. The ICO’s DPIA guidance names artificial intelligence as an example of innovative technology. Using it alongside another high-risk factor, such as sensitive or large-scale data, needs a DPIA. If you are unsure, the ICO recommends doing one.

Transfer checks. If data goes outside the UK, the ICO’s international transfers guidance applies. US transfers can rely on the UK Extension adequacy regulations only where the recipient is certified; otherwise you need safeguards such as the IDTA or the Addendum and a transfer risk assessment.

Data minimisation. Only put in what the task needs.

The ICO’s AI and data protection guidance was last updated on 15 March 2023. The ICO says it is under review after the Data (Use and Access) Act 2025, so it may change.

Which applies to you?

Use this decision list before anyone in the firm pastes client material into ChatGPT.

• If staff use personal Free or Plus accounts, then ban client data in them. Allow only general questions with no client details.
• If you can remove every identifier, then anonymised prompts are lower risk. Strip names, UTRs, addresses, bank details and unusual facts that could identify the client.
• If you need to use real client data, then move to a business plan, sign the provider’s data processing terms, and complete a DPIA first.
• If the data includes special category data, such as health details in a tax return, then treat it as high risk and get advice from your data protection lead.
• If a client has asked you not to use AI, then record it and respect it.

The checklist before you use ChatGPT with client work

1. Write an AI policy. Say which tools and plans are allowed. See our AI policy template.
2. Pick the plan. Use a business plan, not personal accounts, for any client work.
3. Check the contract. Confirm Article 28 processor terms are in place.
4. Check where data goes. Record whether it leaves the UK and which safeguard applies.
5. Do a DPIA. Record the risks and the controls you chose.
6. Update your privacy notice. Tell clients you use AI providers as processors.
7. Anonymise by default. Remove identifiers even on a business plan.
8. Review every output. Check figures and references against primary sources.
9. Train staff. Explain the rules and the risk of prompt injection, which the NCSC says can make a model reveal confidential information.
10. Plan for mistakes. Treat client data pasted into the wrong tool as a possible breach.

A personal data breach that risks people’s rights must be reported to the ICO within 72 hours. See data breach response for accountants.

What firms should do now

Audit which AI tools staff already use, as many will have personal accounts. Then set the policy, choose an approved tool and train the team. For more, read AI data privacy for accountants, GDPR compliance for accounting firms and our review of ChatGPT for accountants.

Frequently asked questions

Can accountants use ChatGPT with client data?

Not in a personal account. ICAEW warns that client data in tools such as ChatGPT is very likely to breach confidentiality. A business plan with processor terms, a DPIA and anonymised prompts reduces the risk.

Does ChatGPT train on what I type?

OpenAI says Free, Plus and Pro chats can be used for training by default unless you turn it off. It says Business, Enterprise, Edu and API data is not used for training by default.

Do I need a DPIA to use ChatGPT in my firm?

Often, yes. The ICO names AI as innovative technology, and combined with another high-risk factor, such as sensitive data, a DPIA is required. The ICO recommends one if you are unsure.

Is pasting client data into ChatGPT a data breach?

It can be. If personal data is disclosed without a lawful basis or proper safeguards, treat it as a possible breach and assess whether to report it to the ICO within 72 hours.