The Law Society of New South Wales updated its AI court protocols tracker on 14 May 2026. The tracker now covers nine separate AI frameworks across NSW and federal jurisdictions. If you have not looked at it recently, the growth since its first iteration is striking. Not in a reassuring way.
Nine protocols. One jurisdiction tracker. And that is before you add Victoria, Queensland, and the other states, each developing their own guidance on roughly their own timetables.
For the past year or so, the standard professional response to AI court protocols has been something like: familiarise yourself with the Federal Court's practice note, keep an eye on your home jurisdiction, and you are broadly across it. That posture made sense when the field was thin. It does not make sense now.
Each of the nine frameworks the Law Society is tracking has its own disclosure threshold, its own confidentiality warnings, and its own working definition of what counts as AI-assisted work; those definitions do not always agree. The Federal Court's GPN-AI is precise about verification obligations and the implied undertaking risk when confidential material enters an AI tool. NSW Supreme Court guidance approaches the same territory differently. The District Court, the Local Court, and NCAT each sit somewhere else on the spectrum again.
The practical consequence is this: "we checked the Federal Court practice note" is no longer a sufficient answer for a multi-jurisdiction litigator. The right question is which protocol governs which proceeding, and whether the firm's AI workflow is defensible under that specific framework, not under the one they reviewed six months ago.
There is a temptation to read this as an awareness gap. Practitioners just need to read more carefully, track updates, stay current. That framing undersells the difficulty.
Awareness is necessary but not the hard part. The hard part is that the answer to "does this protocol require disclosure?" depends on the forum, the stage of proceedings, the type of document, and the tool used. For a litigator running matters across the Federal Court, NSW Supreme, and NCAT simultaneously, which is not an unusual position for a commercial disputes practice, those are different answers in the same week, sometimes the same day. No individual practitioner holds that matrix reliably in their head, and the matrix is not static.
The Law Society's tracker is a genuine public service. It consolidates what would otherwise require monitoring nine separate court websites, practice note portals, and guidance pages. But a tracker tells you what the rules are. It does not tell you whether your research workflow is compliant under each of them.
Underneath the procedural variation, the frameworks converge on something consistent: you must be able to account for AI-generated content before it reaches the court. That means knowing what tool was used, being able to verify that cited authorities exist and say what you say they say, and understanding whether confidential material was exposed in the process.
That last obligation is where generic AI tools create the most exposure. Most general-purpose AI systems process prompts through external infrastructure. Entering client materials, opposing party documents, or compulsory process material into those systems raises implied undertaking questions that several of the protocols now address explicitly. The confidentiality provisions are not footnotes. They are operational constraints on which tools practitioners can use at the matter level.
For research specifically, the compliance floor across all nine frameworks is effectively the same: your sources must be real, and you must be able to show they are. A fluent AI summary of a legal position is not a compliant research output if you cannot trace it to an actual authority. This is not a high bar in principle. It becomes an impossible bar for tools that generate plausible-sounding citations that do not resolve.
This is where we think the category conversation needs to shift. Verified, source-linked Australian law is not a premium feature of a research tool. It is a prerequisite for a research tool that a practitioner can actually use in court-adjacent work under these frameworks.
Habeas scans over 300,000 Australian cases and pieces of legislation, with results grounded in a closed dataset of legitimate Australian legal sources: verifiable and traceable, never hallucinated. That architecture is not a differentiator on a feature sheet. It is the design response to exactly the compliance environment the Law Society's tracker now documents. Research that used to take a full morning can be completed in minutes, and every output points back to a real source that survives the verification step the protocols require.
Nine protocols today. More to come. The firms that build jurisdiction-aware, traceable research into their default workflow now are the ones who will answer the court's question without scrambling when it is asked.
If you want to see how Habeas handles multi-jurisdiction research, book a demo at habeas.ai.
The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.
Hero image: Melinda Gimpel on Unsplash
