Guides

Can staff put client data into ChatGPT or Copilot?

Checked 9 November 2026 against the pages linked in the text. This is a reading of those public sources for a buyer. It is not legal advice, and it is not your regulator's rule unless the paragraph says so.

No. Not as a standing permission. If the text relates to an identified or identifiable living person, it is personal data, and UK GDPR Article 6 says processing is lawful only where at least one lawful basis applies. Article 5 still applies on top: fairness, a specified purpose, minimisation, accuracy, storage limitation, security, and the ability to show your working. ChatGPT and Copilot are not named in the statute. The same test applies to both, and to any other tool.

Is a client name or a client email personal data?

Yes, when it relates to a living person who can be identified. Article 4(1) defines personal data as any information relating to an identified or identifiable natural person, and it gives a name as one of the identifiers. A payroll figure, a home address, a health detail or a tax reference tied to a person is the same kind of information. A company name, with no person attached, is not what Article 4(1) describes. A named buyer at that company usually is.

Pasting the text into a tool is processing. Article 4(2) includes use and disclosure by transmission. The firm that decides why the data is going in is the controller. If the tool provider processes it on the firm's behalf, that provider is a processor, and Article 28 applies.

What has to be true before anyone pastes it in?

First, a lawful basis under Article 6. The list is consent, a contract with the person, a legal obligation, vital interests, a public task, a recognised legitimate interest, and legitimate interests. The article says processing is lawful only if and to the extent that at least one applies. Which one fits is the firm's decision. Consent has a strict definition in Article 4. Legitimate interests has a balancing test in Article 6(1)(f), and public authorities cannot rely on it when performing their tasks.

Second, the purpose limit in Article 5(1)(b). Data collected for a client matter is not a free stock of prompts. Further processing has to be compatible with the purpose the controller collected it for. Third, minimisation in Article 5(1)(c): adequate, relevant and limited to what is necessary. A whole file is the wrong paste if a short extract, or no personal data at all, would do the task. Fourth, security in Article 32. The controller and the processor must put in place technical and organisational measures appropriate to the risk. Fifth, accountability in Article 5(2): the controller has to be able to demonstrate the rest.

If a provider is your processor, Article 28(3) says the processing must be governed by a contract or other legal act. That contract sets out the subject matter, duration, nature and purpose, the type of personal data, the categories of people, and your obligations and rights. It must say the processor acts only on your documented instructions. Article 28(9) says the contract is in writing, and electronic form counts. The statute does not tell you whether a particular vendor's terms meet that list. Someone at the firm has to read the terms.

Does the brand of tool change the statute?

The articles do not mention ChatGPT or Copilot. What changes in practice is who receives the data, where it is stored, and whether you still control it. Vendor settings and terms change, so read the contract you would actually sign.

For solicitors, the SRA warning notice of 17 August 2026, Misuse of AI, adds a professional duty on top of the statute. It says you remain accountable for the work whether or not AI helped prepare it. It says both free and paid tools may store, retain or use what you enter, depending on the provider's terms, settings and technical design. It says client information should only be entered where appropriate contractual, technical and organisational safeguards are in place. The notice also reports the Upper Tribunal's comment, in UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), that putting client letters into an open tool such as ChatGPT places that information on the internet in the public domain. That sentence is the SRA's account of the tribunal, cited here so you can read the notice. It is aimed at people the SRA regulates. A business outside that remit still has Articles 5, 6, 28 and 32.

The Information Commissioner's Office publishes guidance on AI and data protection. This article does not summarise that guidance. The sentences above are taken from the UK GDPR text on legislation.gov.uk, read on 9 November 2026, and from the SRA notice read the same day.

A decision the firm can write down

Article 5(2) is the practical test. A firm that can answer these in writing is further along than a firm that hopes the tool is "enterprise". Which basis in Article 6 are you using, for which people? What is the purpose, and what must staff not use the data for? What is the smallest extract that will do? Who is the processor, and where is the Article 28 contract? What did you do about security under Article 32? Who checks the output before it is used, given that the statute does not move responsibility onto the tool?

Schools in England have a plainer recommendation on top. The DfE guidance, updated 12 August 2025, says personal data must be protected under data protection law and recommends that personal data is not used in generative AI tools. RICS-regulated firms have a stricter upload rule in their own standard: private and confidential data stays out unless there is express written consent in advance and a satisfactory risk check. That standard is on the RICS page.

If you want a session that walks a team through these questions on its own files, that is the workshop, at the prices published there. The session does not clear a file for you. The basis, the contract and the instruction are the firm's.