You can’t use AI to check AI: what Australian courts now expect from legal research
Two of Australia’s biggest Supreme Courts now say it outright. If generative AI helped write it, a human has to check it against the original source, and running it past another AI tool doesn’t count. Here’s what that means for lawyers choosing legal analytics tools.
The rule in plain sight
Most of the talk about AI in Australian courts has been about disclosure: do you have to tell the court you used it? The more important rule is quieter, and it’s now settled in Australia’s two biggest states.
The Supreme Court of NSW’s Practice Note SC Gen 23, in force since 3 February 2025, requires practitioners to check that every citation and authority in an AI-assisted document exists and is accurate and relevant. And it says that verification “cannot be completed by using a generative AI program”.
The Supreme Court of Victoria’s Practice Note SC Gen 25, in force since 14 May 2026, goes further. At paragraph 7.4(e): “One AI tool cannot be used to confirm the content generated by another AI tool.” Asking a tool to confirm its own sources isn’t enough either.
Put simply: you can’t use AI to check AI.
AI practice notes across Australia are heading the same way
The Federal Court’s Generative AI Practice Note (GPN-AI), issued on 16 April 2026, treats a lawyer whose name is on a filed document as having personally confirmed that the authorities cited exist and support the propositions they’re cited for. The Fair Work Commission’s guidance note, in force since 20 October 2026, asks users to personally check every fact, case, statute and quote, and requires lawyers and paid agents to hyperlink the case law they cite.
The details vary from forum to forum. The direction doesn’t. Responsibility for what’s filed sits with the person who files it, and checking means going back to the primary source.
The consequences are no longer hypothetical. A Victorian solicitor’s practising certificate was varied in 2025 after AI-related conduct: for two years he can practise only as a supervised employee solicitor. The College of Law, citing a public hallucination tracker, counted 73 Australian court cases involving AI hallucinations as at June 2026.
Why AI hallucination in legal research is a design problem
Generative AI writes by predicting the most likely next words. A citation that looks right, with a plausible party name, a believable year and a familiar report series, is exactly what a prediction engine is built to produce. That’s why fabricated cases look so convincing, and why a tool can’t reliably catch its own mistakes.
So the courts’ position isn’t anti-technology. It reflects how these tools work. If an answer was generated rather than taken from a source, someone has to rebuild the chain from answer to authority by hand.
That’s where the real cost sits. A tool that saves an hour of drafting but adds an hour of checking hasn’t saved anything. It has moved the risk onto the lawyer whose name is on the document.
The detail lawyers should notice
Victoria’s practice note includes a helpful line. Users may verify AI content by checking “original source material housed within a GenAI tool”.
In other words, the court cares less about which tool you used than about whether you looked at the real thing: the actual judgment, the actual section. A tool that puts the primary source one click away makes your verification duty fast. A tool that gives you fluent prose and a list of citations to chase makes it slow.
Five questions to ask before you rely on any legal analytics tool
- Can I open the source? Every proposition should link to the judgment or provision it relies on, not to a summary of it.
- Is it Australian law, and current? Case law analytics are only as good as the jurisdiction and currency of the material underneath them.
- Was the answer generated or derived? Ask whether the output is predicted text or the result of reasoning over actual legislation and case law.
- Where does my data go? Client material and privileged information should stay in Australia, under your control.
- Could I show my working to a judge? If you couldn’t explain how the tool reached its answer, you can’t stand behind it.
How we approach it at JurisTechne
We built JurisTechne for exactly this standard. Our Strategy Intelligence Reports are anchored to legislation and case law, not language predictions. There’s no generative guesswork: every insight is traced to real legislation and case law, so it’s 0% hallucination by design. Our data is sourced under permission and never leaves the jurisdiction. You can see how our case law analytics work.
That doesn’t replace your professional duty to verify. Nothing should. It means that when you check, the source is already in front of you.
What’s next
Expect more of this, not less. Federal Parliament’s Joint Select Committee on Artificial Intelligence is due to report by 30 November 2026, and more courts and tribunals will update their rules as practice settles. The firms that do well won’t be the ones that use the most AI. They’ll be the ones that can show where every answer came from.
If you’d like to see what a source-traceable report looks like on your own matter, see how it works.
This article is general information, not legal advice. Check the current practice notes for each court you appear in.
