Murray (Wamba Wemba) v Victoria: a Federal Court of Australia indemnity-costs order after AI-fabricated citations

The problem

Generative AI, and AI-assisted research tools generally, can return fluent, authoritative-looking citations to documents that do not exist. When a lawyer files that output without checking it against the underlying sources, false authorities enter the court record. This case is one of the first in Australia to impose a costs sanction for that failure, and a rare one where the fabrications appear to have come not from a public chatbot but from a junior solicitor’s use of the everyday Google Scholar search tool while working remotely (source).

What was built

There is no deployment to celebrate here; this is an AI-governance honest-negative. Massar Briggs Law acted for the applicant in a native title determination, Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria, in the Federal Court of Australia, decided by Justice Bernard Murphy with reasons published on 2 July 2025 (source). The firm filed a document titled the “Applicant’s Summary of Native Title Claim Group and Decision Making Process”, containing footnotes referencing anthropological and historical reports (source). First Nations Legal & Research Services (FNLRS), tasked with producing the footnoted documents, “concluded that most of the cited documents did not exist, and that others existed but were incorrectly cited” (source). As the judgment records, “FNLRS described the citations as ‘fabricated’, which I understood as a reference to the tendency of Generative AI to ‘fabricate’ or ‘hallucinate’ information that looks accurate and reliable but that is not based in fact” (source).

The evidence was that an inexperienced junior solicitor, working on a restricted practising certificate and out of the office without access to the physical or electronic copies of the cited documents, “used the Google Scholar search tool” and took the first result as the citation (source). When concerns were later raised, she attempted to replicate her searches but “Google Scholar produced different search results” and she could not explain why (source). The court found: “It seems likely that the false document citations arose through Generative AI, possibly associated with Google Scholar” (source). The principal solicitor, Jason Briggs, accepted it was an error to allow the collaborative work to be performed remotely and described the failure to ensure the junior solicitor’s work was checked as “an oversight error” (source).

The outcome

The order: indemnity-basis costs against the firm personally. Justice Murphy ordered that “the solicitors for the applicants, Massar Briggs, shall personally pay the costs of the respondents, on an indemnity basis, incurred through the firm’s use of artificial intelligence in the preparation of documents served on the respondents” (source).

The figure that was not fixed: no dollar amount. No sum was set; the court directed that “such costs shall be paid forthwith upon their being taxed or agreed” (source). The court located the fault centrally in the failure to verify the tool’s output rather than in the use of AI itself (source).

The referral that was declined. The court declined a disciplinary referral: “But I do not consider the use of AI in this case means that it is appropriate to refer the solicitors’ conduct to the Victorian Legal Services Board” (source). That distinguishes the earlier Dayal [2024] FedCFamC2F 1166, where a solicitor who filed AI-generated fictional authorities was referred to the Board; here the solicitors had apologised and the court accepted their regret as genuine (source).

Two independent Australian newsrooms reported the outcome firsthand. Lawyers Weekly (Naomi Neilson) wrote that “Massar Briggs Law was ordered by the Federal Court of Australia to personally pay costs for submitting a native title summary document that contained citations that were either incorrect or did not exist” (source). Information Age, the Australian Computer Society’s publication, reported that Justice Murphy “ordered Melbourne-based firm Massar Briggs Law to pay costs for filing two documents containing citations which were made up or incorrectly included by a generative AI (GenAI) tool” (source).

How this was verified

Checked live on 2026-08-12. The origin is a published Federal Court of Australia judgment (Murphy J, [2025] FCA 731, 22 April 2025, reasons published 2 July 2025), read against the AustLII text and byte-tied to organic Wayback capture 20250819013335 (local sha1-b32 3NQNXB3VRWU4VZGJDRYMRKEH6NQXVK64 = CDX digest) (source). Every quoted line above was re-read against that judgment on the live check date. The order, its indemnity basis, the absence of any fixed sum, and the declined referral were restated firsthand by two independent newsrooms, Lawyers Weekly (7 July 2025) and Information Age / ACS (2 July 2025) (source).

The honest limit: the court did not name ChatGPT or any specific model or vendor. It found only that the false citations “seem likely” to have arisen through Generative AI “possibly associated with Google Scholar”; the identified tool is the Google Scholar search function, and the fault sanctioned is the failure to check and verify its output (source). No monetary figure is claimed because none was fixed, and no sum should be inferred; any later taxed or agreed quantum would be a separate document. This page carries the in-queue pending badge, not a green verified badge.

Related case files

The same failure-to-verify pattern, sanctioned in the United States, is Gauthier v Goodyear, where the remedy paired a fixed fine with a mandatory generative-AI ethics course rather than costs alone; it shows how differently two courts can price the same conduct. For a case where the referral this court declined was instead made, see Prososki v Regan, in which the Nebraska Supreme Court struck the brief and referred the attorney for discipline. And for another order that made the offending firm personally pay for AI-fabricated authorities, see OTG v Ottogi, a Rule 11 sanction in the District of New Jersey.

Sources

Tier 1 = the court stating its own order (primary); Tier 2 = independent reputable press naming the parties. All URLs checked live on 2026-08-12.

1. Federal Court of Australia, “Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731” (Murphy J), decided 2025-04-22, reasons published 2025-07-02 (Tier 1, primary court judgment; byte-tied to organic Wayback capture 20250819013335). https://www.austlii.edu.au/cgi-bin/viewdoc/au/cases/cth/FCA/2025/731.html
2. Lawyers Weekly, “Melbourne firm submits AI-fabricated citations” (Naomi Neilson), 2025-07-07 (Tier 2, independent legal-trade press naming the parties and quoting the judgment firsthand). https://www.lawyersweekly.com.au/sme-law/42496-melbourne-firm-busted-using-ai-fabricated-citations
3. Information Age (Australian Computer Society), “Melbourne law firm caught using fake AI citations,” 2025-07-02 (Tier 2, independent technology press restating the order firsthand). https://ia.acs.org.au/article/2025/melbourne-law-firm-caught-using-fake-ai-citations.html