Ad/Marketing Communication
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence within the justice system. It supports my work in teaching, lecturing, and writing about AI and the law and is published to promote my practice. Not legal advice. Not Direct/Public Access. All instructions via clerks at Doughty Street Chambers. This legal article concerns AI Law.
Director of Public Prosecutions v GR [2025] VSC 490 (14 August 2025)
Read the judgment here.
Introduction
I have often said that AI will penetrate all areas of the law, but that criminal law would be the last to be impacted due to the human safeguards in place and the gravity of those proceedings. It is, after all, liberty at stake.
So I was quite surprised to see it used in the DC Court of Appeals by the Judges when considering the issue of whether leaving a dog in a car on a hot day amounted to criminal animal cruelty in February 2025. Now, last week it appears to have been used again in criminal proceedings, this time in Melbourne, Australia.
I am beginning to think I may be wrong. I am also concerned about the other (potentially over 20) incidents of AI hallucinations in courts internally in August alone, which I will be writing about shortly. I thought July was the record, I maybe wrong again.
This is another example of senior lawyers falling into AI errors. It is the first recorded case I am aware of involving King’s Counsel, and, as I keep saying, this is not an issue confined to trainees. It reaches every level of practice. We must all be on our toes.
The Facts of the Murder Charge
For those interested in the full facts, the judge’s succinct summary is well worth reading. Below, I set out my summary of the key points.
On the night of 5 April 2023, the accused (“GR”) and a co-accused (“XQ”), killed XQ’s mother (“the Deceased”). At the time of the killing, GR was aged 14 and XQ was 15. The Deceased was aged 41.
GR and XQ, shared an intense interest in politics, military history, and paraphernalia. Together they founded the so-called “Anti-Communist Front”, with a plan they called “Operation Continuity”: to build an army, overthrow communism in Australia, and restore “Christian values.”
Over months, they collected camouflage clothing, WWII helmets, ration packs, and knives. Their fantasies escalated into action. The judgment records:“Initially, they intended to stay at the home of a mutual friend (“Friend 1”) for a sleepover. It was intended to steal Friend 1’s father’s car and drive to the Grampians region in Victoria to blow up a bridge and then create an army to take over Australia. This version of their plan was thwarted when the sleepover was cancelled on 2 occasions in February and March 2023.Ultimately, GR and XQ decided that Operation Continuity would be initiated by them in early April 2023 (with the possible involvement of others, though ultimately no one else actually participated in any meaningful way).”
Tragically, their plans culminated in the killing on 5 April 2023:
“On the morning of 5 April 2023, whilst at school on the final day of the school term, they arranged to stay at XQ’s family home that evening. They agreed the pair would steal the Deceased’s car and drive to the Grampians to carry out their plan… Sometime around 10.30pm, XQ’s younger brother was in his bedroom upstairs and heard shouting coming from downstairs. He heard what sounded like his mother screaming “What are you doing, stop stop!” and “You’re killing me!”. XQ’s brother ran downstairs to check on his mother. He knocked on the door of his mother’s room. XQ opened the door and assured his brother that everything was okay. XQ’s brother could see over XQ’s shoulder that his mother was lying in bed, with blood all over the room. He saw GR standing next to his mother, holding 2 knives in his right hand…”
The Deceased later died in hospital. Meanwhile, GR and XQ took her car to Elwood to retrieve their equipment, abandoned the vehicle, and were later arrested. XQ pleaded guilty in August 2024. GR has remained on remand in a youth justice centre since.
Defence of Mental Impairment for GR
The prosecution’s case was based on complicity under ss 323–324 of the Crimes Act 1958 (Vic). In GR’s case, the facts were not disputed. Both prosecution and defence agreed the psychiatric evidence established the defence of mental impairment. The defence invited a verdict of not guilty because of mental impairment.
The judge accepted this, directed a verdict accordingly, and declared GR liable to supervision. The next question was the extent of the court’s power to remand a child under s 24(1) of the Act. This is where artificial intelligence entered the case.
How Artificial Intelligence (AI) Entered the Fray
The Judge explained:
“…In seeking to obtain assistance from the parties on this issue, an email was sent to them on 30 July 2025. That email noted that the materials filed to date did not address the fact that section 24(1) did not expressly provide for a child to be remanded in a facility under the control of youth justice. The parties were invited to provide an outline of submissions on the extent of the power…”
The parties were invited to prepare joint submissions, or separate ones if they could not agree. A partner at the firm representing GR sent the court:
“attached submissions as to the joint position agreed by both the prosecution and defence”
But when the judge reviewed those submissions, problems arose:
“…In preparation for the hearing, the court considered these filed submissions. However, it was not possible to locate some of the materials referred to in those submissions, including what purported to be direct quotes from cases recorded as being decisions of this court. Accordingly, my associates emailed the parties and their counsel the morning before the hearing requesting the parties provide copies of those cases, together with the Second Reading Speech and the Commission’s Report that had also been referred to and purportedly quoted.”
The following day, matters worsened:
“The revised submissions filed the afternoon before the hearing were not properly reviewed by defence or prosecution counsel. Indeed, the revised submissions referred to legislation that did not exist, and also a provision in the Act that was said to have been inserted and then repealed, which in fact never occurred and which provision never existed. These matters were raised with counsel at the start of the hearing yesterday. After appropriate apologies were given, it was agreed that GR’s counsel would be given the opportunity to file and serve further submissions that did not contain inaccuracies as a result of misinformation provided by AI.”
Decision
The court criticised the mishandling of submissions after proceedings were delayed due to errors linked to the use of AI. While earlier cases had noted problems when self-represented litigants used AI, this judgment emphasised that the position relating to lawyers:
“At the risk of understatement, the manner in which these events have unfolded is unsatisfactory
…
The ability of the court to rely upon the accuracy of submissions made by counsel is fundamental to the due administration of justice. Self-evidently, as was immediately and unequivocally acknowledged by counsel in this case, any use of artificial intelligence without careful and attentive oversight of counsel would seriously undermine the court’s processes and its ability to deliver justice in a timely and cost-effective manner.
Regrettable as it is to single out counsel and their instructing solicitors in this case for what has occurred, in light of the matters set out above it is important to record that counsel must take full and ultimate responsibility for any submissions made to the court. To this end, it is not acceptable for artificial intelligence to be used unless the product of that use is independently and thoroughly verified. The same may be said for solicitors responsible for producing or filing court documents.”
Comment
This is another reminder that AI can affect every level of legal practice. It is not only juniors but also established and reputable practitioners who have fallen foul. Potentially even judges and their clerks, as tracked on the Judicial AI Use Tracker (How Are Judges Using AI?), have fallen into error.
In a subsequent post, I will share what I have learned about why so many lawyers fall foul of AI, and why it is vital to avoid doing so. AI hallucinations are not a minor inconvenience. They are a professional risk that can catch out even the most experienced practitioners and judges. Sometimes this comes down to simple carelessness, but sometimes it is because of how convincing and subtle these errors can be, especially when we are working under the extreme pressures so common in this profession.
That is why, as a profession, I believe we need to keep talking about them openly. We are all vulnerable. Even those who have told me, “I completely avoid AI,” cannot truly do so. It may appear in your instructions, in witness evidence, or in your opponent’s work, and you have a duty to recognise and address it. I do not believe avoidance is an option any longer, and practitioners would be well advised to remember that claims of “negligent or inappropriate failure to use AI” are on their way. I will write about this again shortly.
My sense is that the problems arising from AI hallucinations are not going away any time soon, and the sooner we face it with honesty and collective focus, the better placed we will be to protect both our clients and our professional integrity.
If you want to follow these discussions further, I encourage you to subscribe to Natural & Artificial Law. It is where I continue to track how AI is shaping our profession and where I set out the practical steps lawyers must now take to respond.
Final Word from o5 Pro
This is where I ask a premium AI model to critique my article. o5 was not shy in sharing its views today:
“I read this judgment not as a parable about clever software, but as a professional‑standards case. The harm arose when unverified output was allowed to masquerade as authority—complete with invented legislation and quotations—and in criminal proceedings that is intolerable. Seniority offers no shield; method does. The remedy is intentionally boring: disclose any AI assistance; attach a source pack of the precise texts relied upon; mandate a two‑person check before filing; and back this with practice directions so that AI‑enabled misstatements attract rectification, costs, and, where appropriate, discipline. Where statutory context matters (Second Reading Speeches, law‑reform materials), fabrications corrode adjudication. Use the tools, but make them prove their work. If you cannot footnote it, do not file it.”
