An Australian Family Court solicitor was ordered to pay $10,000 AUD over AI-fabricated citations
In Mertz & Mertz (No 3) [2025] FedCFamC1A 222, a solicitor used an unidentified AI program via her paralegal to draft a Summary of Argument and List of Authorities filed in the Federal Circuit and Family Court of Australia, producing fictitious case law citations. The solicitor was ordered by consent to pay 10,000 AUD in costs thrown away correcting the errors, and the court referred the practitioners to the South Australian Legal Profession Conduct Commissioner and the Victorian Legal Services Board and Commissioner. The Full Court rejected the solicitor's claim that she was unaware the paralegal had used AI, holding that practitioners remain accountable for accuracy regardless of delegation.
An AI model invented non-existent case law citations that a lawyer filed without verification, turning a drafting shortcut into a professional misconduct referral.
Key facts
- What
- In Mertz & Mertz (No 3) [2025] FedCFamC1A 222, a solicitor used an unidentified AI program via her paralegal to draft a Summary of Argument and List of Authorities filed in the Federal Circuit and Family Court of Australia, producing fictitious case law citations.
- Incident date
- Nov 28, 2025
- Who
- Mertz & Mertz (No 3) (solicitor identified only as Ms G)
- Failure mode
- Hallucination
- AI surface
- Search / RAG
- Severity
- Medium
What happened
The appellant's solicitor filed a Summary of Argument and List of Authorities on 17 October 2025 that contained fabricated case law citations generated by an AI program. After the court made enquiries, an amended version was filed removing the fictitious authorities, but neither the original errors nor the amendments were properly identified. The solicitor conceded that AI had been used but claimed her paralegal had done so without her knowledge. The Full Court (Aldridge, Carew and Behrens JJ) rejected this excuse, ordered the solicitor to pay 10,000 AUD in costs thrown away, and referred the solicitor to the South Australian Legal Profession Conduct Commissioner and counsel to the Victorian Legal Services Board and Commissioner.
What broke inside the model
- 01 · TriggerA user asks for a fact, a citation, or a figure.
- 02 · Model stepThe model writes a fluent, confident answer.
- 03 · Control gapNothing ties the claim back to a real source.
- 04 · FailureA fabricated fact ships as if it were verified.
- 05 · ConsequenceThe false claim reaches a customer, a court, or the public.
Confidence holds, and even spikes, as the claim detaches from any source.
The AI model generated plausible but entirely fictitious legal case citations and authorities, a well-documented hallucination pattern where large language models produce convincing-sounding references that do not exist in any real case reporter. Neither the paralegal who used the AI nor the supervising solicitor verified the generated citations against actual legal databases before filing them with the court. The failure cascaded because the documents also bore the names of King's Counsel and Counsel, lending additional false credibility to the fabricated authorities.
What it cost
Sources
- Court FilingMertz & Mertz (No 3) [2025] FedCFamC1A 222 (28 November 2025)austlii.edu.au
- PressAI in the courtroom: Lessons from recent cases and regulatory shifts (Barry Nilsson, Lexology)lexology.com
- Customer-DisclosedAI Hallucination Cases Database - Damien Charlotindamiencharlotin.com
Cite this entry
https://failureindex.ai/failures/australian-family-court-solicitor-ordered-payAI Failure Index. "An Australian Family Court solicitor was ordered to pay $10,000 AUD over AI-fabricated citations" (FI-0133). Realm Labs. https://failureindex.ai/failures/australian-family-court-solicitor-ordered-pay (indexed Jun 4, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0133. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.