Are we seeing similar developments overseas?

Well, just last week, Alex, the High Court of Ireland issued a practice note and that has similar obligations.

If people make witness statements or affidavits, they have to include a declaration in the opening paragraph saying they have not used generative artificial intelligence for the purpose of generating content.

You can use it for spell checking and expert reports have a similar requirement and again, require upfront disclosure of when artificial intelligence has been used.

And then over in the UK, there's no single binding disclosure rule yet.

The Civil Justice Council's interim report in 2025 recommended lawyers should retain responsibility for AI-assisted outputs.

And they did propose a framework for mandatory disclosure in certain categories.

And there's a recent UK case that illustrates this sort of supervision problem particularly well, Cork and Smith, which was decided by the High Court of England and Wales in May 2026.

So the privacy policy itself became the evidence that the confidentiality requirement wasn't met then?

That's exactly right.

The court didn't just accept the defendant, Mr. Hepner, who had intended to keep things confidential.

It looked at the objective terms and found that under those terms, he had no reasonable expectation of privacy.

The second ground was the work-product doctrine, which is the U.S. equivalent of aspects of our litigation privilege.

And that covers materials prepared by or at the behest of counsel in anticipation of litigation.

And the court held that those materials failed that test because they were created unilaterally, not by or at the behest of his counsel, so fell outside that privilege.

Let's round this out by considering what the key takeaways are off the back of these three decisions.