AI-fabricated quotes and nonexistent citations in a Supreme Court of Victoria murder case

The problem

In a murder proceeding in the Supreme Court of Victoria in Melbourne, before Justice James Elliott, the defence team filed written submissions that contained generative-AI fabrications, “fabricated quotes from a speech to the state legislature and nonexistent case citations purportedly from the Supreme Court” [source] [Source 1, Tier 2]. Justice Elliott’s associates could not find the cited cases and asked the defence lawyers to provide copies; the cases did not exist [Source 1, Tier 2].

The court’s own written reasons record the same first-hand. Elliott J.’s catchwords list “Misinformation produced by artificial intelligence” [source], “Hallucinated case references” and “Accuracy of quotations and citations not verified” [Source 3, Tier 1], and note that the revised submissions “removed fictitious quotes from what were previously said to be parts of the Second Reading Speech” ([source]) [Source 3, Tier 1]. Counsel of record are stated as “R Nathwani KC with A Beech” [source] [Source 3, Tier 1].

What happened

This is a court-origin AI-hallucination record, not a technology deployment: the defence submissions were prepared with the assistance of generative AI and filed without adequate checking. Per Lawyers Weekly, the defence team, “senior barrister Rishi Nathwani KC and junior Amelia Beech,” “failed to properly check the submissions before filing” [source] [Source 2, Tier 2]. The documents held “references to non-existent case citations and inaccurate parliamentary quotes” ([source]) and were “unsigned by both barristers and solicitors” [Source 2, Tier 2], and even the re-filed corrected submissions “still contained further inaccuracies, including references to non-existent legislation and appeals” [source] [Source 2, Tier 2].

The outcome

King’s Counsel Rishi Nathwani took full responsibility and told the court, “We are deeply sorry and embarrassed for what occurred” [source] [Source 1, Tier 2]. Justice Elliott stated, “At the risk of understatement, the manner in which these events have unfolded is unsatisfactory” ([source]), and that “It is not acceptable for artificial intelligence to be used unless the product of that use is independently and thoroughly verified” [source] [Source 1, Tier 2].

The errors “caused a 24-hour delay” [source] in resolving a case Elliott had hoped to conclude a day earlier [Source 1, Tier 2]. On Thursday 14 August 2025, Elliott ruled that the accused, a 16-year-old who cannot be identified because he is a minor, was “not guilty of murder because of mental impairment” ([source]) over the April 2023 death of a 41-year-old woman in Abbotsford [Source 1, Tier 2]. No fine or formal professional-conduct referral is reported in the reachable sources.

Weakest link (source honesty)

The two turning-point facts, the AI fabrications and Justice Elliott’s rebuke, each rest first on a Tier-1 record: Elliott J.’s own published reasons, Director of Public Prosecutions v GR [2025] VSC 490 [Source 3]. The weakest load-bearing source is the AP wire report as carried by NBC News [Source 1]: it is an independent Tier-2 newsroom, not the primary court document, and it is the only source that carries the “24-hour delay” figure and the verbatim in-court quotes. Those quotes are corroborated in substance by the second independent newsroom (Lawyers Weekly) and by the court’s own reasons, but the exact “24-hour” number appears in that single wire report.

Related case files

• Prososki v. Regan, Nebraska Supreme Court strikes an AI-fabricated brief: an appellate court dismissing an appeal and referring counsel for discipline over fictitious AI citations.
• Ringo v. Colquhoun, Oregon Court of Appeals, $2,000 in sanctions: a per-citation and per-quote sanction schedule for AI-fabricated authorities.
• Torres-Campos v. Muñoz, California Court of Appeal, $5,000 sanction: a US appellate court imposing a monetary sanction where Victoria imposed none.
• Smith v. Farwell, Massachusetts Superior Court, $2,000 sanction: an early trial-court sanction in the same AI-hallucinated-citation pattern.