The Supreme Court of Victoria just formalised again what responsible AI use in law actually looks like in practice.
On 14 May 2026, the Supreme Court of Victoria issued a Practice Note ('the Note') for court users and a set of Judicial Guidelines for judicial officers covering the use of artificial intelligence. Together they represent the most substantive guidance an Australian court has produced on AI to date, and they carry legitimate consequences for practitioners who have been using these tools without active consideration of the underlying responsibilities.
The Practice Note draws a clear and unsurprising line: court users are responsible for the content of their court documents regardless of how those documents were produced. The Note is explicit that filing a document containing inaccuracies -whether AI-generated or otherwise - could result in costs orders, and that legal practitioners who rely on unverified AI outputs [understandably] face the prospect of sanctions. The existence of an AI tool in the workflow does not redistribute that responsibility anywhere.
On privacy, the Note flags that different AI tools handle data in materially different ways, and that practitioners need to understand the privacy implications of the specific platforms they are using before putting confidential client information into them. This has been an under-examined question in most firms' AI adoption processes, and the court is now making clear it expects practitioners to have answered it.
On the judicial side, the Guidelines draw an equally firm boundary. Generative AI cannot be used for judicial decision-making. Court-approved tools may assist judicial officers and court staff in defined supportive roles, like organising case materials, producing summaries and chronologies, legal research assistance, proofreading: but these uses are explicitly not a substitute for reading evidence and submissions or for fact-finding. Chief Justice Richard Niall has been consistent on this point, continually noting that AI is not intended to operate beyond its intended capacity as an aid to judicial decision-making, and the new guidelines formalise that position.
Ultimately, a common assumption across most practice notes i that AI in legal work operates under professional responsibility, and that the standard applied to court documents is unchanged by how they were produced.
That framing directly concerns the way practitioners should evaluate the technologies available on the market. The pivotal question is now whether a tool's outputs are accurate enough in the specific legal context in which it is being used for a practitioner to stand behind it. For Australian practitioners, that context is Australian law: Australian statutes, Australian case law, the regulatory frameworks that apply in Australian courts. A tool trained predominantly on US or UK legal material and deployed in an Australian practice naturally imbues less confidence in the capacity for a practitioner to stand behind outputs.
The Practice Note requires practitioners to implement AI in a way that is consistent with their existing professional obligations, which in practice means being able to answer a few questions they may not have asked yet. See the following for instance: 'what is the privacy posture of the tools being used, and is client data being used to train underlying models? Are the AI outputs being verified against primary sources before they reach a court document, or are they being reviewed in the looser sense of being read and accepted? Is the tool being used built on legal material that is actually relevant to the jurisdiction being practised in?'
We consider Habeas to answer those questions very positively for practitioners. If you want to try it yourself, book a demo at www.habeas.ai.
