Ask ten solicitors whether ChatGPT is safe for legal work and you will get ten different answers, most of them confident and several of them wrong. The confusion is understandable. “ChatGPT” is not one product with one set of rules — it is a family of products with materially different terms, and the differences are precisely the ones that matter to a practising solicitor.

So before asking whether it is safe, ask a narrower question: which version, configured how, holding what data?

The distinction that actually matters

The free consumer version of a general-purpose AI tool and the business or enterprise tier are not the same product wearing different price tags. The critical difference is what happens to what you type.

On consumer tiers, your inputs may be retained and used to improve the underlying model. On business and enterprise tiers, the standard position is that your inputs are not used for training, are held under a commercial data processing agreement, and can be subject to retention controls you set.

For a solicitor, that is not a nice-to-have. It is the difference between a tool you can bring near client information and one you cannot.

The practical test

If you cannot say, in one sentence, what the provider does with the text you paste in, you should not be pasting client information into it. That is the whole rule.

What “safe” means in a professional context

Safety here is not really a technology question. It is a professional obligations question wearing technology clothes. Three things need to hold at once:

1. Confidentiality survives the workflow

Client information must not leave your control in a way you have not sanctioned. That means checking the data processing terms, not the marketing page. It also means being deliberate about what actually needs to go into the tool — a great deal of useful AI work can be done on anonymised or structural material rather than identifiable client detail.

2. A qualified person reviews the output

AI output is a draft, never an answer. The value is in the speed of getting to a reviewable first version, not in skipping the review. Any workflow that quietly removes the solicitor from the loop is not a time-saving; it is a liability generator.

3. You can explain what happened

If you were asked six months later how a document was produced, you should be able to answer. Workflows that leave no trace are difficult to defend and difficult to improve.

Why generic tools disappoint on legal work

Set the confidentiality question aside for a moment, because there is a second reason solicitors try AI and give up: the output is mediocre.

This is almost always a setup problem rather than a capability problem. A general-purpose tool, used cold, with a one-line instruction, produces general-purpose text. It does not know your precedents, your house style, the Irish statutory context, or the way your practice actually handles a file.

The same underlying model, given your templates, your tone, a clear description of the task and the relevant reference material, behaves like a different tool entirely. The gap between “this is useless” and “this saves me an hour a day” is usually configuration, not the model.

A short checklist before you use any AI tool on client work

The short version

ChatGPT is not inherently safe or unsafe for legal work. A properly configured business-tier account, used on appropriate tasks, with solicitor review before anything leaves the office, is a defensible setup. A free consumer account with client names pasted into it is not.

The profession’s direction of travel is clear enough — careful adoption with supervision, not prohibition and not enthusiasm. The Law Society of Ireland’s guidance points the same way.

See where the hours are going

A free 15-minute AI admin audit. We look at your actual week and tell you which tasks are worth automating — and which aren’t.

Book Your Free Audit →