Is pasting a client email into ChatGPT a data breach under UK GDPR?
Pasting a client email into ChatGPT can be a personal data breach under UK GDPR, but it is not automatically one. It depends on what personal data was in the email, which account and settings were used, and whether the data reached someone it should not have. Many incidents turn out to be low risk. Some need reporting to the ICO within 72 hours. The first few hours are for finding out which.
What counts as a personal data breach
UK GDPR defines a personal data breach broadly. It covers personal data that is lost, destroyed, changed, disclosed or accessed without permission. Accidental counts too. Nobody has to act with bad intent.
Pasting client details into an AI tool is a disclosure to an outside company. Whether that disclosure was unauthorised is the real question. If your firm has a contract with the AI provider covering that use, it may be ordinary processing. If a member of staff used a personal free account, it is much harder to argue it was authorised.
Is pasting into ChatGPT a data breach? The factors that decide it
Work through these calmly. Write down the answer to each one as you go.
What was in the text
Start with the content itself. A client's name and a meeting time is one thing. Bank details, National Insurance numbers, health information or a list of many people is another. Special category data, such as health or criminal records, raises the risk sharply.
Which account was used
A business plan with agreed data terms is very different from a personal free account. Settings matter too, such as whether chats can be used to improve the model. Our guide to whether ChatGPT or Copilot trains on business data explains how the account type changes this.
How many people are affected
One person's email address is a small exposure. A spreadsheet of every client is not. Count the people, not the files.
Whether the data can be pulled back
Can the chat be deleted? Was history switched off? Deleting it does not undo the disclosure. It does reduce the ongoing risk, and that matters to your assessment.
Who else may have seen it
Was the chat shared by link? Was it on a shared login? Exposure beyond the AI provider changes the picture.
Rule of thumb: the question is not "was data pasted?" but "is there a risk to the people it describes?" That risk decides whether you report.
The first hours: what to do
Speed helps. Panic does not. Here is a sensible order.
- Thank the person who told you. Staff who own up early are your best early warning. Blame teaches the rest to stay quiet.
- Get the facts. Which tool, which account, what time, and exactly what was pasted. Ask for a copy of the text if it is still available.
- Contain it. Delete the chat. Check the account's data settings. Remove any shared links.
- Tell your data protection lead. If you have a DPO, bring them in now.
- Assess the risk. Use the factors above. Decide how likely harm is, and how serious it would be.
- Decide on reporting. Report to the ICO within 72 hours of becoming aware if the breach is likely to risk people's rights and freedoms. Tell the people affected directly if the risk is high.
- Record it either way. UK GDPR expects you to log every breach, including ones you decide not to report. Write down your reasoning.
The ICO publishes a self-assessment tool and guidance on breach reporting. It is worth reading before you decide. If you are unsure, take advice. Not every incident needs reporting, but getting the decision wrong is the bigger problem.
A worked example
Example: At Ashcombe, Priya pastes a client email into a personal ChatGPT account to tidy the reply. It holds the client's name, phone number and a note about a late payment. One person. No special category data. She deletes the chat within the hour and tells her manager. Ashcombe logs it, notes the low risk, and decides not to report. If the email had held the client's bank details and health notes, the answer could well be different.
Check your other duties
UK GDPR is not the only rule. Solicitors have confidentiality duties to the SRA. Financial advisers answer to the FCA. Client contracts may say who you can share data with. Some cyber insurance policies ask you to tell them about incidents. Check each before you close the matter.
Stopping the next one
Banning AI rarely works. Staff use it on their phones instead, out of sight. A better aim is simple. AI is welcome. Client data stays out of it.
- Write a short AI policy. Say which tools and accounts are approved, and what must never be pasted.
- Get each person to agree to it. A dated record helps when a client or insurer asks.
- Teach redaction. Swap names for labels before pasting. Our guide on how to redact sensitive data before pasting into ChatGPT walks through it.
- Know what tools are in use. You cannot manage what you cannot see.
- Add a check before anything is sent. Training fades. A check on the PC catches the slip on a busy afternoon.
That last point is what we built Helio Guard for. A browser extension on Chrome and Edge checks each paste, message and file upload on listed AI sites. Card numbers, UK bank details, National Insurance numbers and client names are taken out and swapped for stand-ins such as "Client A". The question still goes. Checked on the PC, with fixed rules. We never see what your team types. Windows only for now. It is £1 a person a month, from 5 people, and the pricing page has the detail.
Common questions about pasting data into ChatGPT
Do I have to report every ChatGPT incident to the ICO?
No. Under UK GDPR, a personal data breach only has to be reported to the ICO if it is likely to result in a risk to people's rights and freedoms. Low-risk incidents still need recording in your internal breach log, with your reasoning, so you can show why you decided not to report.
Does deleting the ChatGPT chat undo the breach?
Deleting a ChatGPT chat does not undo the disclosure, because the data has already reached the provider. Deletion does reduce ongoing risk, and it is a sensible containment step. Record when the chat was deleted and what settings the account had, as both count when assessing the incident.
Should the member of staff be disciplined?
Disciplining staff for an honest mistake usually backfires. If people fear punishment, they stop reporting slips, and unreported breaches are far more dangerous than reported ones. Focus on clear policy, training and checks. Deliberate or repeated misuse of client data is a separate matter for your normal HR process.
Is pasting anonymised data into ChatGPT a breach?
Pasting truly anonymised data into ChatGPT is not a personal data breach, because UK GDPR only covers information about identifiable people. Be careful though. Removing a name is not always enough if a job title, address or unusual detail still points to one person. Strip anything that could identify them.
Related: How to stop staff pasting client data into ChatGPT without banning AI, How to stop employees sharing sensitive data with AI tools, not just ChatGPT, Is it safe to paste bank statements into ChatGPT? What to take out first