Relativity recently hosted an AI Masterclass in Melbourne, bringing together general counsel, senior in-house leaders, law firm partners, and legal technology executives for a full day of candid conversation about where AI in ANZ legal stands, what courts and regulators now expect, and what the strongest legal teams are building to lead over the next three to five years. Across four sessions, here's what emerged.
Where AI in ANZ Legal Stands
The opening session set the tone: less about what AI could theoretically do, and more about what's being adopted right now.
Two live disputes framed the conversation. A recent ruling from the U.S. District Court for the Northern District of California, in Schulte v. LinkedIn, didn't make headlines for approving AI-assisted review, since that's already settled law at this point. The real question the court was probing was how the workflow was built, how it was disclosed, and whether it was proportionate to the matter – a shift worth paying attention to.
Gina O’Neill, associate director at Ashurst Perkins Coie, observed that whilst Australia hasn’t had a headline case addressing the use of AI, the courts are becoming increasingly active in shaping expectations. This is reflected in the growing number of practice notes, guidance materials, judicial seminars, and recent court orders.
For instance, see recent orders from Blue Sky and then Virgin. Both are identical.
Karan Mehta, head of legal technology and client services at Allens, brought the conversation back to the practical question sitting underneath every governance debate: not whether AI can be used, but how.
"The real question isn't whether we can use AI; it's how we use it responsibly and at scale," he said. "That's where the practice notes and frameworks earn their keep." To bridge the transformation gap, we need much more than a powerful model. It requires quality data, integration with existing systems, clear governance, change management, and people who understand both the technology and the business problem they're trying to solve.
Dr. Mitchell Adams, senior lecturer at Swinburne University of Technology, argued none of it matters until the fundamentals are in place: "This all comes back to education," he said. "Before you can talk about compliance, you have to actually understand how the AI works."
That emphasis on foundational understanding is also the premise behind Mitchell's own book, GenAI for Legal Practice, which introduces an AI Fluency Framework for practitioners integrating AI tools without compromising professional standards. You can download the open-access book here.
What Courts and Regulators Now Expect
Dr Susan Bennett, founder and executive director of InfoGovANZ, was clear about the baseline: "This is fundamental; firms should already be able to identify where and how AI is being used and under what controls." She likened AI traceability to the long-standing discipline of checking a draft affidavit for accuracy and pointed out that lawyers remain responsible for the work before a court, including where AI has been used. Existing duties of competence, diligence, professional independence, and supervision continue to apply. Like company directors, she argued, lawyers are "on notice" when it comes to AI use.
Craig Macaulay, data scientist at Phi Finney McDonald, named the practical version of that oversight: "cognitive controls," the discipline of reviewing data and prompts to make sure they don't contravene Harman (confidentiality) obligations or reach into discovery data they shouldn't touch. He was candid about how easy it is to be lulled by AI output.
"The output is intoxicatingly nice," he said, and typically mostly right, but it's that remaining gap that can bring a matter undone under trial-level time pressure.
It's a theme both panellists kept circling back to: human judgment, human oversight, human perspective. Not a checkbox, but the things that hold a governance framework together.
These expectations are no longer theoretical, either. The Federal Court's AI practice note (GPN-AI), released in April 2026, sits alongside a growing body of state Supreme Court level guidance, and courts are already using live matters to test how AI-assisted work should be handled.
Justice Michael Lee, a Federal Court judge, is already putting that into practice: he's using a hundred-million-dollar class action against McDonald's as a testing ground for artificial intelligence, urging lawyers to get imaginative about cutting massive case costs.
Craig flagged the bind that puts firms in hesitant to disclose AI-generated work today, but potentially exposed tomorrow for failing to use AI for efficiency at all. Susan pointed out that clients using AI to improve efficiency and reduce costs are likely to expect their law firms to use AI appropriately to improve efficiency and reduce legal costs as well.
Susan pointed to the Australian Government's six Essential Practices as the clearest practical guidance for organisations implementing AI governance. Firms should know who is accountable for AI, maintain an AI register, adequately resource governance, and build AI literacy across their organisations. AI governance should not simply be left to the technology team.
(The 6 Government AI Governance Pillars are the benchmark Susan referenced. Relativity's AI Principles, which guide how we build AI into our platform, are built to work hand in hand with the framework above.)
Craig's closing caution was the necessary counterweight: too many guardrails, applied without thought, can be just as damaging as too few. If every firm ends up making the same decisions because they're all constrained by the same rigid controls, the industry loses the differentiator that matters most, critical thinking.
"Human in the loop," he warned, can become a superficial concept if it isn't genuinely built into how a team works.
What High-Performing Legal Teams Are Building Now
The agent moment, much discussed and much hyped, is already giving way to something more embedded, and the panel was candid about what that demands.
Dr. Susan Bennett set the tone early: agentic AI is still at an early stage of enterprise development, particularly in legal workflows. But agentic capabilities are already emerging in organisations, which is exactly why a conversation about safe agentic AI matters now. Aurelie Jacquet, CEO at Ethical AI Consulting, brought the international vantage point to the room.
“What we're seeing play out here in ANZ mirrors what's already happening in other jurisdictions; the gap isn't in ambition, it's in the governance maturing fast enough to keep pace with adoption," she said.
Jenna Chambers, legal counsel at Phi Finney McDonald, pointed out the need to consider data privacy obligations in relation to AI. Best practice includes considering risks and required privacy disclosures where agentic AI may lead to automated decision making. High-performing legal teams understand the risks associated with data processed by AI technology, how that technology processes that data, and consider that risk in how they use AI. The risks for lawyers using AI aren't new as legal professional obligations remain unchanged. High-performing legal teams have policies in place for ensuring privilege is maintained and obligations such as the implied obligation and client confidentiality are extended into the use of AI. Susan put it plainly, proposing it go straight into the agenda: "Risk of agentic AI in vendor products already being used," the thing most legal teams aren't paying enough attention to right now.
Jenna added a warning few catch in time: that a vendor has already thought through the legal and ethical risk of their own product is not a given, and broad terms allowing vendors to use data for product improvement present a risk of inadvertent breach legal professional obligations. Jenna confirmed the need to ensure that contracts with external service providers, including consultants, barristers, and experts, include terms that ensure their use of AI technology does not lead to a breach of the implied obligation, client confidentiality, or a waiver of privilege. High-performing legal teams also understand the need to meet their costs disclosure obligations when using AI to complete their work and are actively thinking about the tension between traditional billable hours pricing models and the rising, unpredictable cost of tokenisation.
Dylan McGrath, senior manager at Ashurst Perkins Coie, was direct about the real capability gap: AI fluency built across the entire workforce, not ring-fenced inside legal tech teams, alongside a growing need for specialists who can convert institutional knowledge into playbooks the whole firm can use.
That's the thread running through all of it: as AI becomes more embedded, human judgment doesn't become less important. It becomes more important, and harder to demonstrate.
Susan closed the loop on hiring: "You can't rely solely on a vendor's own representation of its product." Having the right people, with clear authority and responsibility for both AI governance and its ongoing management, is critical. The capability to adapt to fast-paced change, collaborate across domain expertise, and accountability are crucial, not simply someone's job title. Her prediction: firms that assume they already have the right capabilities in the right roles may come to learn otherwise.
Introducing claiR: Conversational AI, Built for Lawyers
The Masterclass closed with an announcement that speaks directly to everything the day had explored.
Relativity claiR is conversational AI built specifically for lawyers, not a general-purpose tool adapted for legal use, and not a feature bolted onto existing software. It's designed from the ground up for the complexity, precision, and accountability that define legal work, built on 25 years of Relativity's understanding of how legal professionals operate and what's at stake when things go wrong.
claiR brings AI into the workflows legal teams already rely on, under the kind of real oversight courts and regulators require. For ANZ teams navigating the shift from pilot to practice, it's what responsible, embedded AI use looks like in practice.
claiR is currently in limited availability, with general availability expected early 2027.
Graphics for this article were created by Sarah Vachlon.
Recommended for you
