What should our firm do if there is a data breach involving an AI tool?

Treat it as a breach first and an AI problem second. Your existing incident procedure applies: contain the exposure, establish what data was involved and whose, decide whether it is reportable to the Information Commissioner's Office and whether the client needs to be told, and record what you did and why. The AI tool changes where you look and who you ask, not the obligations you are under.

Establish what actually happened before you label it

Most suspected AI incidents in law firms fall into a small number of patterns, and they are not equally serious.

• A fee earner pasted client material into a general consumer tool that the firm has not approved and has no contract with.
• Someone in the firm could see a matter they should not have seen, because access inside the approved tool was set too widely.
• Material from one client's matter appeared in work done on another client's matter, usually because files were loaded into the wrong place.
• The supplier told you about an incident on their side.
• A document went out containing something that should not have been in it, and AI was involved in producing it.

The first step is factual. Which documents, which clients, which individuals, what categories of personal data, how long the exposure lasted, and who had or may have had access. Ask the person involved directly and early. People hide these things when they expect to be punished, and a day lost at the start is the most expensive day of the whole process.

Where an approved tool is involved, you should be able to see the record: the conversation, the documents loaded, the account that did it, the time. If your supplier cannot tell you that, you have learned something important about them.

Confidentiality and data protection are two separate questions

They overlap but they are not the same, and a firm that only thinks about one of them gets the response wrong.

Your professional duty of confidentiality covers client information whether or not it identifies a living individual. A corporate client's deal terms exposed to a third party is a confidentiality problem even if no personal data is involved at all. That duty, and the Solicitors Regulation Authority's expectations about reporting serious matters, run on their own track.

Data protection law applies where personal data is involved. If there has been a personal data breach, you need to assess the risk to the individuals concerned and decide whether it must be reported to the Information Commissioner's Office, which expects notification without undue delay and within a short deadline where the threshold is met. If the risk to individuals is high, they may need to be told as well. Your data protection lead should be making that assessment on the facts, not the IT supplier and not the fee earner who made the mistake.

Record the decision either way. A documented decision not to report is defensible; silence is not.

Telling the client

The question most firms struggle with is whether to tell the client when the exposure looks trivial. The honest test is simple: would the client expect to be told, and would they be entitled to feel misled if they found out later from someone else?

If the answer to either is yes, tell them. Say what happened in plain terms, what data was involved, what you have done about it, and what you are changing. Do not pad it with reassurance you cannot support. Clients generally react worse to a late or evasive account than to the incident itself, and the Legal Ombudsman sees plenty of complaints where the handling did more damage than the original error.

If you are reporting to the Information Commissioner's Office, assume the client will find out. Tell them first.

Fix the conditions, not just the incident

After the immediate response, look at what made the breach possible. In practice it is usually one of four things.

1. No approved tool. People use what they can reach. If the firm has nothing sanctioned, staff will use consumer products on client papers and will not tell you.
2. Access set by seniority rather than involvement. If everyone can see everything, every mistake is a bigger mistake. Access should follow the matters a person is actually working on.
3. Weak account security. Unverified accounts, shared logins and no second factor turn a small problem into an unbounded one.
4. No record. If you cannot reconstruct what was done, you cannot scope the breach, and you end up over-reporting out of caution or under-reporting out of ignorance.

Also check your supplier contract now rather than during the next incident. You want to know how quickly they must tell you about an incident on their side, what detail they will give you, where your data sits, and who they use underneath them.

Finally, make the internal consequence of reporting an error proportionate. A firm that treats every mistake as a disciplinary matter will not hear about the next one.

Where Alesis fits

Alesis is an AI assistant for UK law firms, made by Alesis.ai Limited and used through the web browser, with one conversation for a matter. The firm's information is held in the UK, processed only in the UK and the EU, kept apart from every other firm, and never used to train anything for anyone else. Inside a firm, people see only the matters they are on, and seniority alone grants no view. Every account proves its email address, and can add two-factor authentication and passkeys.