AI-generated errors delay justice: Lessons from a Victorian case

A murder trial in the Supreme Court of Victoria faced disruption when submissions filed by the Senior Counsel for the accused contained fake quotes and non-existent judgments that had been generated by Artificial Intelligence (AI).

Relevant documents:
Director of Public Prosecutions v GR
Crimes (Mental Impairment and Unfitness to be Tried) Act 1997

Disclaimer: The information provided in the articles in this section are of a general nature and should not be construed as specific advice or relied upon in lieu of appropriate professional advice. Whilst LEAP Legal Software uses commercially reasonably efforts to ensure the information in these articles are up to date at the time of publication, LEAP does not warrant their accuracy, currency or completeness and excludes all loss or damage howsoever arising (including through negligence) in connection with the information contained in these articles.

Quick takeaways

• Every AI output must be independently and thoroughly verified before being relied upon in legal proceedings

• AI will not replace lawyers, but lawyers who fail to understand its limitations risk being left behind

• Clear guidelines around responsible AI use in legal practice are becoming increasingly necessary as courts set expectations

• The most effective approach combines AI efficiency with human oversight, keeping a lawyer in the loop at every stage

The incident in detail

The case, Director of Public Prosecutions v GR, saw the submissions presented to the Court following a request by Justice Elliott on 30 July 2025. Justice Elliot had asked that if parties could reach an agreement, they should provide an outline of joint submissions by 6 August 2025.

At 4:54pm on 5 August 2025, submissions were filed representing a ‘joint position agreed by both the prosecution and defence.’ These documents contained references to Supreme Court judgments, including direct quotations, a Second Reading Speech for the Crimes (Mental Impairment and Unfitness to be Tried) Bill 1997 (‘the second reading speech’) and the Victorian Law Reform Commission’s final report that followed a 2014 review of the Crimes (Mental Impairment and Unfitness to be Tried) Act 1997(Vic) (‘the Act’).

However, issues arose when Justice Elliott’s associates attempted to verify the references. They could not locate the cited Supreme Court judgements and subsequently asked the parties to provide copies of the judgments, along with the Second Reading Speech and the Act.

Following the request, Senior Counsel for the accused emailed the Court to admit the errors. They stated that the citations were incorrect and did not actually exist. Senior Counsel acknowledged that the cases referred to were invalid and took full responsibility for the mistake.

The explanation offered was that initially the citations were checked and found accurate, leading to a mistaken assumption that subsequent cases and citations were also correct.

A valuable lesson

This case serves as an example of AI’s power, but also its pitfalls. It demonstrates that even highly experienced criminal lawyers are adopting these tools, yet it stands as a critical reminder: AI can make mistakes. It cannot replace the essential work of a lawyer independently verifying every detail.

Justice Elliott’s comment in his judgment captures this issue perfectly:

“It is not acceptable for artificial intelligence to be used unless the product of that use is independently and thoroughly verified”.

This incident proves that while AI will not replace lawyers, legal professionals who don’t adapt to or understand the limitations of new technology will eventually get left behind.

The role of technology in law

We can compare this shift to when computers first replaced typewriters. That transition dramatically improved the efficiency for lawyers and many other professions. AI will do the same for traditional legal work. It is a tool that will change how lawyers work and create, rather than just speeding up existing tasks.

This incident also highlights the growing need for clear guidelines around the responsible use of AI in legal practice. As courts develop expectations for AI-generated content, it is crucial for lawyers to verify AI outputs before relying on them in important matters.

Keeping the human-in-the-loop

At LEAP Legal Software, the generative AI research assistant (LawY) is designed to help lawyers interpret legal queries and surface relevant, jurisdiction-specific resources. To support accuracy, LawY’s hybrid approach allows AI-research answers to be verified through a lawyer-in-the-loop process, combining the efficiency of AI with careful human oversight.