Hi everyone, it’s Allison Benson here from Thoughts of A Strata Lawyer and Benson Legal.

Today, I want to talk about generative AI. It’s a bit of a bugbear. If this sounds like a rant, I do apologise.

I am seeing it more and more in my practice area, where clients are relying upon generative AI to produce both briefs to us to try and seek advice but also to generate by-laws and sometimes to generate letters to and from other people within their owners corporations.

It’s a really dangerous practice. Any use of AI really should be checked.

The courts are actually finding it’s now being used for the preparation of submissions and evidence. The New South Wales Civil Administrative Tribunal, NCAT, and pretty much all of the courts across Australia, and a lot of other common law countries have made procedural directions about the use of generative AI.

And you know, what is the big fuss?

I’m actually going to quote Chief Justice Bell because I have a less highbrow way of saying this, but I think that his words are quite a good summary. This was in a case, and it was quite a complex case, about property and constructive trusts. The case was May v Costaras [2025] NSWCA 178. Chief Justice Belt, a paragraph 2 of the decision, said:

“I wish to add some brief observations to what his Honour has said at [49] below in relation to the respondent’s use of Generative AI in the preparation of her oral submissions in this Court. In so doing, I make no personal criticism of the respondent who was self-represented and doing her best to defend her interests. But the respondent’s use of Generative AI in the present case highlights the dilemma presented by the availability of that technology, especially where it is used by a person without legal training or otherwise not familiar with or unable to discern both the relevance and accuracy of what Generative AI may produce. I make no personal criticism of the respondent who was self represented and doing her best to defend her interests.”

Now, let’s break that down because we are seeing this. I’m getting briefs that have been written, 30 plus pages of briefs sometimes, and I’m getting by-laws and letters that have been written in my practice (but not by my firm!) by generative AI.

Breaking it down, what the Chief Justice is saying is you need to be really careful if you’re using generative AI to make sure that you understand what it is you’re actually saying because in that case, my way of summarising what the Chief Justice said was, that unfortunately, the respondent had just thrown the documents into a generative AI system and come up with word salad. And the result, well it just didn’t make sense.

If you are using generative AI, double check what has actually been submitted and used to generate the AI response. Maybe the letter or report does not make sense. If you don’t have a base understanding of the legal concepts, then you’re at risk of the word salad result happening.

The AI generated response might sound great, because all the cool words, all the highlighted words are there in it. But the context of those words in each decision/letter/report/response actually matters.

When you get the word salad happening, and these words are just thrown all together randomly, which does happen with these large language models AI uses, then it really doesn’t make sense, and instead of actually assisting, it just shows that whoever has generated the document didn’t understand the concept in the first place. So that’s that’s one really big issue.

Also, it can add to the costs of a matter quite significantly.

We’ve been asked several times to review what are very clearly generative AI produced by-laws, and the by-laws have been lacking, let’s just say. They’ve summarised ideas and concepts, and they haven’t been particular enough for the scheme, or sometimes they’ve just left things completely out that you would normally see for works by laws. For instance, making the liability for the works that have been done by the lot owner, that of the lot owner, and alleviating the owners corporation from the responsibility for any damage caused by those works. Please be very, very careful.

The other thing that’s been coming through in the cases, and there’s actually a substantial amount of case law coming through now, is that there are hallucinations, and so sometimes the cases that have been cited just don’t exist. They can’t be verified. In other cases, textbook references, or cases themselves (precedents) have been used, but have been completely misstated. And sometimes it’s both. Which is a bit strange, but sometimes in the cases, in the decided cases, the courts and the tribunals have gone through, tried to find the case law, read the case law and when they can’t find the proposition that has been cited, they have quickly realised that the words used have been taken out of context and misstated by the person who was trying to rely upon them.

Now, again, what does that mean for you in practice?

One, it can run up costs.

Two, if you’re doing it in proceedings, you’ll actually be in breach of most courts and tribunals procedural directions.

In one case where a generative AI was used, the Tribunal said we’re not going to give it much weight as evidence. There was an appeal to the NCAT Appeal Panel and the Appeal Panel said, actually, that’s correct, you shouldn’t give it much evidence because that’s in accordance with our procedural directions. And so the evidence that was produced using the generative AI was given very little weight.

Three, there have been findings as to credit upon the people that have been seeking to rely upon the generative AI.

And one big thing that I think is, we’re going to start seeing is potentially costs orders against people who do this repeatedly, or even in a significant way and then run up costs in the proceedings.

So, if you’re going to use it, be very, very careful, verify the sources, make sure you understand what it is saying.

Obviously, this is my personal opinion. Please seek advice if you’re going to use it. If something is complicated, it is okay to go to a lawyer and to ask for assistance. I, for instance, did not know much at all with regards to tax law and, you know, ran off and sought assistance on tax provisions. There is no reason why people need or should know everything about every single subject matter. So, if in doubt, seek assistance.

Hopefully this has helped everybody.

Some of the non-strata specific cases where Courts and Tribunals have considered the use of generative AI are:

Access Super Audit Pty Ltd v Seobi Marketing Pty Ltd t/as Online Marketing Gurus [2026] NSWCATAP 186 – where the Appeal Panel noted that evidence generated by AI and sourced from internet searches was given minimal weight due to concerns about reliability in accordance with the Tribunal’s Procedural Direction on the use of Generative AI.

Bottrill v Graham (No 2) [2025] NSWDC 221 – where the Second Defendant had used Generative AI to prepare submissions, submissions that contained misstated and non-existent references, in breach of the Court’s Practice Note.

PH v Children’s Court of New South Wales [2026] NSWSC 589 – where the plaintiff used AI-generated scripts that, although legally phrased, were inconsistent, inaccurate, and nonsensical. This ultimately detracted from her case and negatively impacted proceedings due to the inefficiency and confusion introduced by this material.

Sharma v SafeWork NSW [2026] NSWIRComm 1056 – where in the Industrial Relations Commission, the plaintiff filed submissions prepared with generative AI but failed to meet verification requirements under the relevant practice note. The plaintiff’s lack of understanding of the requirements of the practice note was noted during the hearing, and his submissions were deemed of limited assistance.

Next blog, I’ll go into the strata cases where generative AI has been used and what was said in relation to it.