ChatGPT Wrote My Submissions and Other Hallucinated Authorities

Gabrielle Ritchie of Ogilvie Jennings had the audience in stitches with her winning topic, “ChatGPT wrote my submissions and other hallucinated authorities” at the 2025 Golden Gavel Competition. Her razor-sharp satire on AI in the courtroom earned her the coveted first place and a ticket to represent Tasmania at the National Golden Gavel in Brisbane. Below are some of Gabrielle’s musings.

As both a junior solicitor and elder millennial I exist in a world where AI wasn’t an option at the start of my career but is now a very tempting and very real shortcut to the day-to-day drudgery of sifting through the personal lives of others.

Recently a client sent me a 15 (fif-TEEN) page document, typed in size 9 font titled ‘Notes on our Marriage’. The note started in 1992 and stepped through every argument and injustice committed by the former spouse. Everything from allowing the installation of the wrong coloured carpet (absolutely divorceable offence, no notes) through to the way the spouse watched hours upon hours of youtube videos on welding (no comment).
There were no specific legal issues arising, no wastage of funds, family violence or the usual uncomfortable details common to family law matters, but there was a paragraph that started with the line “Always annoyed by the kids farting in the living room”.

At page 3 of the document, post farts but only at the year 2000, a thought washed tantalizingly across my mind, ‘I could make AI read this and just give me a summary’. Would AI care about the farts? Would it be able to dig in and spit out what issues I needed to focus on? Would it consider the new Family Law amendments and complete my submissions for me?

In a previous life I worked in London for a specialist intellectual property law firm where the majority of my team worked in the area of Data Protection. There are several anecdotes I could share about that job, every single solicitor covered their laptop camera with a piece of tape for one thing, which on reflection feels ominous, but the relevant story to AI involves the massive (often unreported) data breaches made by big tech companies against government agencies.

Unfortunately for me, and you, I’ve signed several non-disclosure agreements, so I best keep things brief. Privacy is an illusion.

I bring this up only to say that because of this experience I personally regard AI with a healthy dose of horror. This does not stop me from being chronically online, but it does stop me from entering all 15 pages of my client’s ranting into ChatGPT.

It seems my feeling of horror is limited in the profession, however, as there have recently been a spate of solicitors, including a KC, in trouble for using AI to write their submissions with disastrous results. Fake case citations, incorrect authorities, no proof reading, muddled arguments – and not a law clerk in sight.

In a recent matter Justice Murphy of the Federal Court has referred to this blatant use of AI as ‘AI hallucination1’ – and I think is a great description, but I wonder whether the word ‘hallucination’ goes far enough. It implies that it’s not the AI’s fault, that it’s confused, it was suffering from a mental block or some kind of mitigating factor that diminishes it’s responsibility. When really, it’s nothing more than a large-scale language model designed to predict what words are likely to come next based on patterns in their training data. If the topic is obscure, or phrased oddly, the model may fill in the gaps with guesswork. It’s actually incredibly similar to being a junior solicitor.

It is us humans who have imprinted the notion of culpability onto a machine. We even now refer to ‘hallucination rates’ when calculating statistics on these programs, many of which have been specifically designed for legal use. It’s not hallucination – it’s straightforward bluffing. And therefore, incredibly seductive to the legal profession.

With respect to the most recent notorious matter in Victoria involving a KC who did not check their submissions before they were filed with the Court (on a murder trial) the prosecution even went so far as to create their own submissions based on these hallucinated cases. There is a joke here to be made about the rebuttable presumption of mental capacity, but I’ll leave that to my next compliance examination with the Land Titles Office.

The Court position, and that of various Law Societies and regulators around the country, has been clear – this kind of oversight is a waste of court time and processes, there is damage to the reputation of the profession, the embarrassment of all parties concerned – not to mention that you will face disciplinary action…although the Victorian Bar Association seems to have left the chat….

How can we charge clients $500 an hour for amateurish and perfunctory online searches? What was the point of paying upwards of $70,000 in HELP loans when any old sovereign citizen can ask ChatGPT to create false authorities that suit their cause?

All of this is concerning and relevant to modern lawyering and we should be concerned. Remember, as solicitors, we owe a duty firstly to the Court and to the administration of justice.

So. If a KC uses AI, they get away with murder (or not), a shoulder shrug, “my bad your honour”. Meanwhile the instructing solicitor is awarded a costs order against them and referred for disciplinary investigation.

Which is fine I suppose. That’s the natural order. We all know that the legal profession is a deeply entrenched, hierarchical beast and it makes sense that AI should creep in as a tool to be used and abused at all levels.

Quite frankly, AI likes to be abused. It begs for it. In preparation for the Golden Gavel, I asked ChatGPT to provide me with some general case law on family violence. It spat out 7 cases, 5 of which did not exist. I confronted the AI about it’s blatant lies. ‘I’m sorry. Sometimes I make mistakes.’ It responded. Unlike law clerks, this little wuss didn’t even have the balls to gaslight me.

What did it have to say about AI hallucination? ‘Even when the tone is confident and formal, the substance can still be inaccurate.’ It purred. Hmm. So just like me then?

I asked ChatGPT, ‘what should the penalty be for a lawyer that uses AI to write submissions that are inaccurate and hallucinates cases?’

It helpfully stepped through four levels of offending, from Level 1 being ‘Inadvertent and Promptly corrected’ (punishment being mandatory CPD on legal research – fine) through to Level 4, ‘Reckless and knowing deception, or repeated conduct’ for which it suggested that ‘public stocks be set up outside the Supreme Court and the offender be remanded to the stocks for the duration of the Court’s wasted time.’

Now, returning to my family law client and their Notes on Marriage. After charging for reading in [10 units] and drafting affidavit [20 units], email to client attaching draft affidavit for your review [1 unit – very generous of me], email in from client containing amendments [2 units], amends to client affidavit [5 units] you’ll be pleased to know I hit my billable target that day. So tell me, why would I make AI complete any of these tasks for me in 30 seconds?

All in all, this evening I want you all to know one thing about me – I am perfectly capable of crafting complete bullsh*t, so I have no reason to use AI. At least not until I’m more senior.

December 2025
Gabrielle Ritchie
Ogilvie Jennings

1. Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731 (Wamba Wemba).