AI and vulnerable clients: extra care in family and criminal matters
By the ChamberOne team 5 minute read
A woman leaving an abusive partner has a phone her partner can read. A young man on bail has a mother who answers his messages. A parent in a custody dispute sends a late-night message that says more about their state of mind than their case. These clients are the reason family and criminal lawyers are cautious with new technology, and the caution is well placed. AI tools can still help, if the limits are set before the first use.
Key points
- Treat vulnerability broadly, and look at your regulator's guidance as the starting point.
- Remove identifying details before using any tool, and avoid consumer tools for safety-sensitive facts.
- Tell clients if AI is used and check they understand, especially where capacity is in question.
- Keep a clear, quick route to a human for distress, and never leave a chatbot to answer it.
- Limit approved uses to tasks you can check, and keep strategy and safety decisions with people.
Who counts as vulnerable, and why it changes the rules
Vulnerability in legal practice is wider than most people assume. It includes clients at risk of harm from a partner or family member, clients with limited capacity, children and young people, people in custody, people with mental health conditions, people who cannot read or write comfortably in the language of the proceedings, and anyone in acute distress. Often several overlap.
For these clients, a mistake costs more. A leaked address can put someone in danger. A misread instruction can affect liberty or contact with a child. So a tool that is acceptable for a commercial contract review may be unacceptable here, or acceptable only in a narrower form. Look at the guidance from your regulator or bar on vulnerable clients and treat it as the starting point.
Privacy and physical safety come first
Think about who can see what, on which device. Before using any technology with a vulnerable client, ask how they prefer to be contacted, whether others have access to their phone or email, and what is safe to say in a message. Agree a code word or alternative channel if needed.
Then apply the same thinking to AI tools. Addresses, children's names and school details, medical information and the location of a refuge should not be entered into a tool unless you are sure where the data goes and who can read it. In many tasks you can remove identifying details and still get the benefit.
- Remove or replace names, addresses and dates of birth before using a tool for drafting or summarising.
- Do not use free consumer tools for anything involving safety-sensitive facts.
- Check whether your tool keeps a history that others in your office could see.
- Make sure transcripts, recordings and summaries are stored with the same restrictions as the file.
- Remove or replace names, addresses and dates of birth
- Check whether the tool keeps a history others could see
- Store transcripts and summaries with the same restrictions as the file
- Use free consumer tools for safety-sensitive facts
- Enter a refuge location or children's details unless sure where the data goes
Capacity, consent and understanding
A client who does not understand how a tool is used cannot meaningfully agree to it. Explain in plain words if you use AI on their matter, what for, and what you do to check it, and note the conversation. If a client's capacity is in question, take particular care: the decision to share their information with a technology provider is also yours to think about, not only theirs.
Tools can also help clients understand. Plain-language explanations of a process, reviewed and sent by you, can be valuable for someone who finds legal letters frightening. They are not a replacement for a conversation with a person who can see whether the message has landed.
Limit what the tools are used for
A sensible approach is to define a short list of approved uses for these matters and treat everything else as not permitted until reviewed. The aim is not to ban technology but to avoid drift, where each person quietly adopts different tools.
- Approved: chronologies from documents with identifiers removed, checklists for hearings, plain-language explanations that you review and send yourself.
- Approved with care: summaries of long disclosure or case papers, in a tool the firm has vetted.
- Not approved: client-facing chatbots giving advice, automatic replies to messages about distress or safety, and any tool that cannot say where data is held.
- Not approved: decisions about bail, contact, safety planning or plea, which need a person who knows the client.
- Chronologies from documents with identifiers removed
- Checklists for hearings
- Plain-language explanations you review and send yourself
- With care: summaries in a tool the firm has vetted
- Client-facing chatbots giving advice
- Automatic replies to messages about distress or safety
- Any tool that cannot say where data is held
- Decisions about bail, contact, safety planning or plea
Keep a human contact for distress
A client who is frightened at midnight needs a reply from someone who can recognise a crisis. An automated assistant cannot judge whether a short message signals danger. If your firm uses a chatbot or automated intake, make sure that every route has a visible way to reach a person, and that messages mentioning harm, self-harm or immediate risk are seen by a human quickly.
Tell clients which numbers to call for emergencies, and which for the firm. Be honest about when you will respond. A clear boundary is more reassuring than a promise you cannot keep.
Special points in criminal matters
Criminal clients may be in custody, with limited access to phones or the internet, and every word of their instructions may be significant. Privilege and confidentiality are central, and any tool must be assessed with that in mind. Material from prosecution disclosure often includes sensitive information about witnesses and other people, and putting it into an unvetted tool can create real harm.
Accuracy matters even more. A summary that misstates a time or a place could mislead the whole defence. Read the underlying material for anything that will be relied on in court, and never let a tool's version replace your own reading of the key documents.
Special points in family matters
Family cases hold information about children, finances and relationships, often across two households that are in conflict. The risk of material reaching the wrong person is high. Be careful with shared devices, joint email accounts and cloud services that sync across a household.
Tools can help organise a long exchange of messages into a timeline, or compare two versions of a financial statement. Do not rely on them to assess a child's best interests, a risk assessment or the credibility of an allegation. These are matters for professional judgement and for the court.
A short policy for vulnerable-client matters
Write a one-page rule for the practice. It should say which tools are approved and for what, that identifiers are removed by default, that clients are told about AI use, that a human reads every client-facing message about safety or distress, and who the firm's lead on vulnerable clients is. Review it each year, and after any incident.
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
AI and access to justice: an honest assessment
Where AI can lower the cost of routine legal help, where it cannot replace advice, and who bears the risk when it is wrong.
5 minute readClient portals: what self-service does for clients and the firm
Which questions a portal answers, what to keep out of it, and how it reduces routine calls.
5 minute readAI chatbots on law firm websites: benefits, risks and rules
A chatbot can help with intake, but it can also say too much. Decide its limits before it goes live.
4 minute read