Could Your AI Chat Become Exhibit A? Five Legal Risks

By Edward Hazzan, Commercial Disputes, Croft Solicitors

Your collective groan was audible across the Cotswolds. “For goodness’ sake, not another article about AI.” I know. Sorry.

This is not an article predicting that artificial intelligence will transform civilisation, abolish the working week or finally make printers operate reliably (they never will).

I use AI enthusiastically. It can organise information, accelerate routine work and remove a remarkable amount of drudgery. But it can also preserve the exchange you hoped nobody would see, disclose information you were obliged to protect, and produce a polished answer that is completely wrong.

From a disputes lawyer’s perspective, the central problem is that an exchange with an AI chatbot feels disposable and private. Legally, it may be neither. Here are five risks worth thinking about.

Your casual AI chat may become “Exhibit A”

In litigation, parties typically have to disclose documents relevant to the dispute, including documents which damage their own case.

“Document” has a broad meaning. It is not limited to signed contracts and formal letters. Any electronic data or communication in any medium or format could count as a “document” if it is relevant to the issues in the case.

AI prompts and outputs are not exempt. If they are relevant to the dispute and fall within the disclosure obligations of a case, they can become evidence just like an email conversation with a colleague.

That matters because people are remarkably candid with chatbots.

A user might ask: “How can we get out of this contract without paying?”; “Can we create a paper trail to justify this decision?”; “What reasons can we give for terminating this supplier?”; “How likely is the other side to find out?”

Your carefully drafted letter setting out your polished legal position may be rather less persuasive, or at worst completely undermined, when read alongside the original prompt.

A High Court judge once gave me a useful rule of thumb: if you wouldn’t want it read out in court, think twice before writing it down. AI prompts are no exception.

The cheapest legal adviser may become the most expensive mistake

An increasing number of parties to legal disputes appear to be using AI chatbots in place of a solicitor or other qualified adviser.

From one perspective, I can understand the attraction: the machine writes reams of correspondence in perfect legalese, for pennies. Meanwhile the opponent pays a real adviser to read it. The chatbot user can play this game all day, all while draining the other side’s resources.

But there is a problem. A chatbot does not have conduct of the dispute. It does not know which facts the client has omitted from the prompt, which points should be conceded, how the opponent is likely to react or whether an apparently clever argument is self-defeating. It cannot accept professional responsibility for the advice, and it does not bear the consequences if its tone or content causes the dispute to escalate.

My colleagues and I now regularly see correspondence in which the legal content is stated with complete confidence, but which is unfocussed and incoherent in substance. These communications may cite irrelevant contractual terms, allege wrongdoing without a proper basis, and ultimately turn a negotiable disagreement into a matter of principle.

In the long run, this strategy often turns out to be more expensive. An opponent faced with a stream of aggressive or incoherent chatbot-generated correspondence may reasonably conclude that further informal engagement is pointless and that the only realistic course is to instruct solicitors and issue formal proceedings. They’ll be seeking their costs for all this too.

The sensible chatbot user may now instruct a solicitor of their own. That lawyer has to get up to speed on several exchanges after positions have hardened. The cheap option has now become the expensive one. Moreover, the chatbot’s exaggerated account of the user’s legal position may now have costs consequences. Whoops.

AI can help a party organise a chronology, identify questions to ask and understand unfamiliar terminology. It cannot reliably replace an adviser who is able to assess evidential weaknesses, explain procedural risk and, importantly, tell a client when an argument should be dropped entirely.

Uploading legal advice may put privilege at risk

Legal professional privilege protects confidential communications between a lawyer and client made for the purpose of giving or receiving legal advice. It allows clients to speak frankly to their lawyers without ordinarily having to disclose those discussions to an opponent. It’s central to the proper operation of our legal system.

That privilege can be lost if the document or communication ceases to be confidential to the lawyer and the client.

So imagine that a business receives advice from its solicitor, pastes it into a free chatbot and says “Summarise this and suggest a response.”

The advice has now been supplied to an external entity: the chatbot provider. Whether privilege has been lost will depend on the circumstances, but the act itself creates an argument that need never have arisen.

The position is not yet settled in England and Wales. Much may turn on whether confidentiality was preserved, what the LLM provider was entitled to do with the material, who could access it, and whether the user had a reasonable expectation of privacy.

Consumer and free-tier AI services deserve particular care because their terms frequently permit prompts and uploaded material to be retained or used to improve the service. Paid and enterprise products can offer stronger contractual and technical controls, but a subscription fee is not a confidentiality guarantee. (A judge will assume you read all that small print.)

The safest course is not to upload legal advice or lawyer communications unless the tool has been approved for that purpose and appropriate confidentiality protections have been checked.

A hallucination can become a representation

AI systems sometimes produce information which is plausible, specific and entirely wrong.

That is commonly called a “hallucination”, a charming term for fabricated information presented with confidence. (I can think of less charming alternatives.)

In the context of a legal dispute, “the chatbot made it up” is not going to be an answer.

Where false AI-generated material induces someone to enter into a contract, that can lead to misrepresentation or negligent misstatement claims. In other circumstances, there may be contractual or regulatory consequences, loss of credibility and, in proceedings, procedural or costs sanctions.

A particular danger with AI is that the risk is increased by the quality of the presentation. A weak human draft often looks weak, and it would be harder for a claimant to argue that they relied upon something that was obviously cobbled together. But when an AI is wrong, it is often wrong in a way that is beautifully structured and presented with headings and citations, making it more plausible. And the chatbot does it all with the serene confidence of somebody who doesn’t have to worry that they’ll ever be cross-examined.

Bottom line: every important factual or legal proposition needs to be checked against a reliable primary source. Don’t be beguiled by the chatbot’s fluency.

Confidential and personal information may travel further than expected

I find that the more context you give an AI tool, the better its answer will be. That encourages users to upload whole contracts, employee files, customer complaints, correspondence and databases.

Doing so can expose personal data relating to customers or employees. Contracts often impose confidentiality obligations, and documents are often supplied for limited purposes or in confidence.

Before using an AI system, a business should know where its data goes and what happens to it. Does the provider retain a copy? Is it transferred outside the UK? Can anyone else access it?

These are not merely questions for the IT department. They concern contractual confidentiality, data protection, employment duties and the preservation of commercially sensitive information.

A sensible AI policy should therefore be practical rather than aspirational. Telling staff to “use AI responsibly” is not much help as the level of trust placed in the systems will vary depending on familiarity with LLMs and their limitations. Staff at all levels need to know which tools are approved for which purposes, what information must not be uploaded and when human review or extra care is required.

Take a mindful approach

None of these risks means that businesses should avoid generative AI. It means they should use it with the same discipline applied to other business tools.

Before entering a prompt, ask:

1. Would I be comfortable having this exchange read aloud in court?
2. Am I sharing privileged, personal or confidential information?
3. Have the important facts been independently checked against sources?
4. Is AI assisting my own judgement or replacing it?

AI is an exceptionally capable assistant. But it is not automatically confidential, it is not always correct and it is not professionally accountable.

This is not an argument against using AI. Far from it. Use it enthusiastically. Just do not let that enthusiasm deprive you of judgement.

This article provides general information only and is not legal advice. Your legal position will depend on your particular circumstances.