The short answer is that I’d rather you didn’t, at least not into a free consumer version, and not without checking with me first. Part of that is a lawyer being precious about his own work, which I’ll admit. But there’s a legal point underneath it, and it’s one most people have never heard of.
I’ve been asked this a few times recently, sometimes directly and sometimes because a client mentions in passing that they “ran it through ChatGPT to see if it made sense”. It’s a reasonable thing to want to do. Legal advice can be dense, and asking a machine to explain it in plain English is a sensible instinct. The problem is what may happen to the advice once it’s in there.
What legal professional privilege is
When you get advice from a solicitor, the communications between you are protected by legal professional privilege. In practice that means if you later end up in a dispute, the other side cannot demand to see what the solicitor told you, even though the ordinary rule in litigation is that relevant documents get disclosed to both sides. A regulator can’t insist on it either. It’s one of the more powerful protections in English law, and it belongs to you, not me. You can give it up; I can’t give it up for you.
The condition attached to privilege is confidentiality. Advice that has been shown to the world isn’t privileged any more, because the thing privilege protects has already gone. So if you send my advice to someone who isn’t under a duty to keep it confidential, you risk losing privilege in it. Depending on the circumstances, you can also lose privilege in related material, though that’s a litigator’s point and I’ll leave it there.
The classic way this happens is a client forwarding advice to a business contact, or to the other side in a negotiation, to strengthen their hand. The AI question is the same problem wearing new clothes.
Why an AI tool might count as a third party
OpenAI’s own help pages say that on its services for individuals, which includes the free version of ChatGPT, it may use your content to train its models unless you opt out, and that a limited number of staff and service providers can access content in certain circumstances (you can read it here). Business and enterprise versions are treated differently: no training by default, and contractual confidentiality terms. Other providers have broadly similar splits between free and paid tiers, though the details vary and change often enough that I wouldn’t rely on my summary of them.
So the question is whether pasting privileged advice into a consumer chatbot amounts to disclosing it to a third party who isn’t bound to keep it confidential. The honest answer is that I don’t know, because as far as I’m aware no English court has decided the point. You could argue that a provider’s privacy policy creates enough of a confidentiality obligation to preserve privilege. You could also argue that a service which reserves the right to train on what you type, and to let its staff read it, is nowhere near a confidential recipient in the sense the law means. If I were acting for your opponent, I’d run the second argument, and I’d ask in disclosure whether you’d uploaded anything.
That last point is something to consider. In litigation each party has to disclose relevant documents, and your chat history with an AI tool is a document. If my advice is sitting in there and privilege has gone, it’s potentially disclosable.
The other risk, which has nothing to do with privilege
Even where privilege isn’t in play (most advice never ends up anywhere near a courtroom), there’s a plain confidentiality problem. My advice to you will often contain your commercial position, your negotiating limits, and personal information about other people: the person you’re in dispute with, the employee you’re exiting, the shareholder you want to buy out. Putting that into a free tool with training turned on is a data protection question as well as a privilege one, and the people named in the document didn’t agree to it.
Yes, I use these tools too
I should say that I use AI tools every day, including to help with the drafting of this article. So I’m not in a strong position to lecture anyone. The difference, and I think it’s a real one rather than a convenient one, is the terms. The tools I use are on paid business terms that don’t train on what I put in, with data processing terms in place, and I’ve had to think about what my firm’s obligations are before using them. I also teach Lawyers how to use AI and have a book called The Lawyers Prompt. The Law Society has published guidance for solicitors on exactly this, and it’s fairly blunt about free consumer tools.
The same logic applies to you. If your business has a paid or enterprise AI service with proper confidentiality and no-training terms, the position is different, and a conversation with me about whether it’s sensible for a particular matter is a short one. If it’s the free version on your phone, please don’t.
What to do instead
If something I’ve sent you isn’t clear, ask me (or your solicitor). A plain English explanation is part of what you’re paying for, and I’d much rather spend ten minutes on the phone than have you interpret my advice through a tool that has never read your file. If you want to use AI to help you draft a reply or think through options, tell me first and I’ll say whether I’m relaxed about it in that instance; often I will be, if the sensitive material is left out.
And if you’ve already done it, don’t panic. Tell me. It may not matter at all, and if it does, it’s far better that I know now than that it comes out later in a witness box.
