UK: data protection law when AI processes client data
By the ChamberOne team 5 minute read
A UK firm that sends client documents to an AI service is processing personal data, often a great deal of it, and often about people who are not its clients: opponents, witnesses, employees, children. UK data protection law applies in the ordinary way. The AI label does not change the questions. It changes how easily the answers get forgotten. This article is general information for managers and data protection leads, not legal advice. Check the current guidance from the UK data protection regulator and take advice on your own arrangements.
Key points
- Start with a data map: what each AI tool receives, where it goes and what the provider does with it.
- A firm is usually a controller of matter data, and an AI provider acting on instructions is usually a processor. Put it in a written contract.
- Check where processing happens and which transfer mechanism applies, and write it down.
- Keep a register, the contracts, any assessment, an audit record and retention rules.
Start from what data goes where
Before any legal analysis, make a plain list. For each AI tool: what data does it receive, in what form, from which matters? Does it include special category data, such as health information, or criminal allegations? Does the provider keep it, for how long and for what purpose? Who are its own subprocessors, and where are they?
Most compliance problems with AI services are not subtle. They arise because nobody asked what the tool did with the data. A simple data map, kept up to date, prevents most of them.
Controller and processor
In general terms, a firm acting for a client usually decides why and how personal data in the matter is used, and so is usually treated as a controller of that data. An AI provider that handles the data only on the firm's instructions, to provide the service, is usually a processor. That relationship should be set out in a written contract with the terms that data protection law expects. Check the regulator's guidance on what those terms are, and have your vendor's terms reviewed.
Things become more complicated if the provider uses the data for its own purposes, such as improving its models. Then it may be acting as a controller for that use, and the firm needs to think hard about whether it can allow it at all. Many firms decide that client material must not be used to train anyone's model, and require that in the contract.
- Acts for a client
- Decides why and how personal data in the matter is used
- Handles data only on the firm's instructions, to provide the service
- Needs a written contract with the terms the law expects
- May be a controller for its own uses, such as improving its models
Lawful basis and transparency
A firm needs a lawful basis for processing personal data, and the right one depends on who the data is about and why it is used. For the firm's own clients, the contract and the firm's legitimate interests are commonly considered. For third parties in a matter, the position is different, and legal professional privilege and other rules may interact with the usual transparency requirements.
Whatever your analysis, check that your privacy notice describes your use of technology accurately and that it is understandable. If clients would be surprised to learn that their documents are processed by an AI service, say so up front.
International transfers
Many AI providers process data outside the UK. Moving personal data to another country needs a mechanism that the law recognises, and the available mechanisms and their conditions change from time to time. Ask each vendor where processing takes place, which of its subprocessors are involved and what transfer mechanism it relies on. Record the answer.
Remember that a provider may host in one place and route requests through another. Ask specifically about where the AI model runs, not only where the data is stored.
Impact assessments
Where processing is likely to result in a high risk to people, a formal assessment of the risks may be needed before it starts. Whether a particular AI use meets that threshold depends on the data and the purpose. Large-scale processing of sensitive data, or new technology used on vulnerable people, are the sort of features the regulator points to. If in doubt, carry out a short assessment anyway, and keep it. It forces the questions that need asking and is useful evidence of care.
Practical records to keep
- 1
A register of AI tools: purpose, provider, data types, location of processing, subprocessors and the date approved.
- 2
The contract or terms for each, with the data protection terms highlighted.
- 3
A note of the transfer mechanism, where data leaves the UK.
- 4
Any impact assessment, with the date and the people who signed it off.
- 5
An audit record showing who used which tool on which matter.
- 6
Your retention rules for AI inputs, outputs and logs, and evidence that they are applied.
- 7
An incident procedure, covering how a suspected breach involving an AI provider would be handled and reported.
Individual rights and AI
People may ask what data the firm holds about them, or ask for it to be corrected or deleted. If AI-generated summaries or outputs are kept in the matter file, they are part of what the firm holds, and you need to be able to find them. Check how your tools store prompts and results, and whether you can delete them when you need to. If a provider keeps copies you cannot reach, that is a gap.
Be careful about decisions made about individuals by automated means. Most law firm uses support a lawyer's decision, and are not decisions in themselves, but a firm acting for an employer on a screening or scoring tool should look at the rules on automated decision making.
Questions to ask a vendor
- Do you act only on our instructions, and will you sign a processor agreement?
- Is any of our data used to train or improve models, now or later?
- Where is data processed and stored, and which subprocessors are involved?
- How long do you keep inputs and outputs, and can we delete them?
- Can we turn AI features off, or choose which provider is used?
- How would you tell us about a breach, and how quickly?
What this does not settle
Data protection is only one of a firm's duties. Confidentiality, privilege and the rules of the firm's regulator all apply as well, and sometimes point in different directions. The guidance is evolving, and the regulator has published material on AI and data protection that you should read directly. Use this article as a checklist of questions, and take specific advice where the facts are complicated.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
Generative AI guidance for lawyers: how six markets compare
How to find what regulators, courts and bar bodies have said in six markets, and the common themes across them.
5 minute readUK: what the SRA and judicial guidance say about AI use
A plain summary of the published positions of the SRA and the judiciary, and what they imply for supervision.
5 minute readEU AI Act: what it means for firms outside the EU with EU clients
How a firm outside the EU should think about the EU AI Act when it works for European clients.
5 minute read