Tech Suite | Legal privilege meets artificial intelligence: What the courts are saying

So Oliver, let's briefly set the scene for any listeners who might be joining us for the first time or who want a quick refresher or who missed the last one.

And if you missed it, go back and have a listen.

It was really good.

What were the core questions we were grappling with last time?

So what we explored really was as lawyers and clients increasingly use AI tools to communicate, generate documents, analyze information, what about legal privilege? And the real question we looked at was does inputting privileged information into an AI model constitute a disclosure that could waive privilege even unintentionally? And we looked at the three main types of privilege solicitor-client, which covers confidential communications made for the purpose of obtaining professional legal advice.

We looked at litigation privilege, which is wider and covers communications and materials prepared for the dominant purpose of action or apprehended proceedings.

And we looked briefly at without prejudice privilege covering confidential settlement communications.

And I think the key point you made, which I do think really landed with our listeners, was around the distinction made between public AI tools and enterprise-grade AI tools.

Can you just quickly recap that distinction? Because I think it's central to the cases we're about to discuss.

Yeah, that's right.

So where we landed was public AI tools being the free versions of ChatGPT or other AI tools that people can sign up to typically have terms of service that allow the AI provider to use the data for training or monitoring in a feedback loop.

And that's typically the default setting.

Sometimes you can opt out of it, but without more, it's the default setting.

And the concern is that if you input your privileged information into one of these tools, you're acting in a way that's inconsistent with the claim of confidentiality.

And that's precisely the test for waiver of privilege.

That's the standard test

today.

On the other hand, you've got the enterprise AI platforms.

These are paid versions, typically, with much greater security controls.

And they don't use data for training and keep it locked down within your cloud environment, again, dependent on settings.

And where we landed last time was that you should essentially treat any AI platform as hostile until proven otherwise and use enterprise grade as the minimum standard for legal work just to protect you on that waiver point.

And that distinction has turned out to be very important, as we'll see from the cases.