The use of artificial intelligence in Bermuda litigation is a growing area of interest that must be navigated carefully by lawyers and their clients.

There are no court rules or formal guidance specific to AI use in Bermuda litigation, but a case here and developments overseas are beginning to show what is expected of parties and their lawyers.

A case decided in Bermuda last year considered what it should realistically cost to search a company’s documents.

In FourWorld v Enstar, shareholders bought out in a merger they had not backed asked the court to decide what their shares were worth. They sought wide-ranging disclosure (also called discovery), the process by which parties to litigation must search for relevant documents and provide them to the other side.

The company objected to its breadth and, assuming a manual review, estimated that producing the documents would take more than 400,000 hours and cost about $100 million.

The shareholders’ expert in electronic discovery estimated $1.5 million to $2.5 million.

In that case, Mr Justice Andrew Martin held that estimates based on manual review were not realistic “in the modern context of commercial litigation”, and that the cost would be “greatly reduced” by electronic discovery “using AI and other forensic software”.

So, a party resisting disclosure on costs grounds will find it harder to rely on manual review estimates. At least for discovery, AI has a place in Bermuda litigation. The harder questions come next, and the answers matter to clients too.

Key questions include:

Where can you trust AI?

Did you check it?

Who else has seen it?

On the first question, document review, the issue in Enstar, is the clearest case for AI.

Survey results published this summer by the judge-led group reviewing England’s disclosure rules show that some practitioners think large human review teams may not always match AI-assisted review on accuracy or cost.

Drafting and research are different. There, AI gives you a useful starting point, fast, but not a document that you can rely on. It misses context, makes mistakes a lawyer with judgment would catch, and sometimes invents things.

What’s left is the hard part: checking it, reworking it and judging whether an argument will actually hold up.

On the second question, a recent English disciplinary case shows the importance of checking AI-generated output.

In August, England’s Solicitors Disciplinary Tribunal decided its first case on AI misuse. The lawyer, defending himself against unrelated charges, cited AI-generated authorities that did not say what he claimed. When this was pointed out, his correction, also drafted with AI, contained fresh AI-related errors.

The tribunal said this alone would have justified striking him off, which it did. He is appealing.

Invented cases are the easy ones to catch. The harder ones are real sources summarised slightly incorrectly, like a quotation that has lost its qualification.

Catching those takes someone who knows the subject. You cannot spot a confident error in something that you do not already understand.

That knowledge comes from doing the work. As AI takes on more of the routine work that juniors once did, clients will expect to see the saving. However, if junior lawyers never do that work, today’s saving comes at the cost of the judgment those same clients will need from them later.

The third question, who else has seen the information, raises risks to confidentiality and privilege.

A UK immigration tribunal dealt with an adviser who had put draft client e-mails, and Home Office decisions on his clients’ cases, into ChatGPT. It observed that this put the information in the public domain, breaching confidentiality and waiving privilege.

That was an observation rather than a ruling on a contested claim, but the risk is real. The practical question is whether the provider can keep or reuse what is uploaded. Clients who paste their lawyer’s advice into a free chatbot to check it run the same risk.

For law firms and businesses, Bermuda’s Personal Information Protection Act applies, too.

Bermuda does not yet have court rules or guidance for lawyers written specifically for AI, though the judiciary’s 2025 annual report said a committee led by Mrs Justice Subair Williams would work on a practice direction.

The Barristers’ Code of Professional Conduct 1981 already requires counsel to be competent and diligent, and to take professional responsibility for all work entrusted to them, however it was produced. Counsel who rely on unchecked AI output risk paying costs personally and facing disciplinary proceedings.

The committee working on the practice direction will face a choice. Trinidad and Tobago now requires court users to disclose when they have used AI in court documents. Conversely, England’s guidance to judges says lawyers using AI responsibly generally need not mention it.

Taken together, Enstar and the English cases point the same way: the courts expect parties to make sensible use of the technology, but the responsibility stays with the lawyer. Anyone relying on AI in a dispute should be ready to answer all three questions.

• Kilian Elkinson is a senior associate in Appleby’s Dispute Resolution practice in Bermuda. A copy of this column can be obtained on the Appleby website at www.applebyglobal.com. This column should not be used as a substitute for professional legal advice. Before proceeding with any matters discussed here, persons are advised to consult with a lawyer