AI Summit 2026: What the Room Decided About the Future of Legal AI

The legal profession has largely moved beyond asking whether artificial intelligence will change legal practice. The harder questions now concern what happens after adoption: how legal AI should be governed, where professional accountability sits, how its value should be measured, and what becomes more valuable as technology makes some forms of legal production faster and less costly.

On 26 August 2026, more than 260 legal leaders, technologists and industry innovators gathered at the Telstra Insight Centre in Sydney for the LexisNexis AI Summit 2026.

They came from across Australia, New Zealand and Papua New Guinea — from law firms, in-house legal teams, government and academia — with a shared interest in a conversation that has rapidly outgrown the question of whether AI belongs in legal practice.

Demand for the event exceeded capacity. More than 500 people had registered. The sessions were full.

That fact alone said something about where the profession is. Not waiting. Not watching. Deciding.

In this article

• Legal AI adoption is no longer the most important measure
• The question has changed from productivity to economics
• Healthcare's lesson for legal AI: start with the problem
• What trustworthy legal AI actually requires
• Why human oversight becomes more important as AI improves
• From experimentation to practice: what scaling actually looks like
• The public sector dimension: governance before speed
• How AI is changing the economics of legal services
• What becomes scarce when legal production becomes cheaper
• What legal leaders should take from AI Summit 2026

Legal AI adoption is no longer the most important measure

Carol Chris, Managing Director Pacific at LexisNexis, opened the AI Summit with a welcome that set the tone for everything that followed. Rather than beginning with a prediction about where artificial intelligence might take the profession, she began with evidence about where legal AI already is.

Drawing on LexisNexis research into AI sentiment among legal professionals across the Asia-Pacific region, she described a profession in which AI has moved rapidly from experimentation towards routine use, and where adoption is no longer, by itself, evidence of successful transformation.

A legal team can give every lawyer access to an AI tool without materially changing the way work is performed. Individual gains need not translate into organisation-wide improvements. And a firm can increase AI usage without being able to answer the more important question: what measurable value has actually been created?

That tension between use and value became one of the recurring themes of the AI Summit. Trust was the other.

Carol's framing was direct:

“Adoption is no longer the milestone that matters. Trust, security, integration, and measurable value are. Getting AI into the hands of your teams was the easy part. Making it dependable enough to run your business on — that is the real work. And it is true for the whole industry.”

The question has changed from productivity to economics

Lindsay O'Connor, General Manager - Content Pacific, who moderated the afternoon's legal economics discussion, gave the shift a particularly useful formulation.

Much of the first phase of the legal AI conversation has centred on productivity, and increasingly, the answer is yes. AI can produce first drafts faster, accelerate research, and summarise large volumes of material.

But productivity is only the beginning of the economic story.

“The more interesting question is what happens to the economics of legal services when the cost, speed, and scale of producing legal work starts to change.”

Once a technology changes the amount of time required to produce legal work, it eventually creates questions about far more than efficiency - the relationship between time and value, pricing models built around units of professional labour, who trains junior practitioners, and who captures the productivity dividend.

Healthcare's lesson for legal AI: start with the problem

One of the most useful perspectives at the AI Summit came from outside the legal profession.

Associate Professor Narinder Singh, Chief of ENT, Head and Neck Surgery at Westmead Hospital and founder of the Society for AI in Medicine, Surgery and Healthcare, presented the Drumbeat AI project: a smartphone-based diagnostic tool designed to detect and triage ear disease among Aboriginal and Torres Strait Islander children in rural and remote communities.

The problem the technology addresses is significant. In some rural and remote Indigenous communities, ear disease prevalence exceeds 40% - the highest in the world. Untreated, it impairs hearing at a critical developmental stage with consequences that extend into education, employment and the justice system.

Drumbeat began not with a technological capability but with a bottleneck: children could be screened, but specialist diagnosis could take considerably longer. AI closed that specific gap.

Associate Professor Singh summarised the principle simply:

“Don't start with AI. Start with the problem.”

He reinforced it with an example from healthcare: an AI system that could detect diabetic eye disease with remarkable accuracy, then failed entirely in real-world deployment across clinics in Thailand because of poor image quality, clinical workflow friction, and patchy internet access. Technically impressive. Practically useless.

"A successful AI project is not one that produces technological innovation. It is one that produces a better outcome."

What trustworthy legal AI actually requires

The question of outcomes leads quickly to the question of accountability.

Dr Owen Bradfield, dual-qualified as a doctor and a lawyer and Chief Medical Officer at the Medical Indemnity Protection Society, approached that issue from the perspective of another high-stakes profession.

His conclusion was particularly relevant to lawyers.

“Accountability primarily and ultimately sits with the end user. Whether that is a lawyer or a medical practitioner. AI doesn't change the standard of care that is expected of a professional. But it introduces new questions. Was it appropriate to use AI in the first instance? And was it appropriate and reasonable to rely on that output?”

Trust is architectural. It is produced by the relationship between the AI model, the material on which it relies, the workflow in which it operates, the guardrails surrounding its actions and the human professional responsible for the final outcome.

A general-purpose model may be extremely capable at language generation while still being poorly suited to a legal task if the underlying information is incomplete, unverifiable or disconnected from the relevant jurisdiction.

Nor does AI remove a lawyer's professional obligations because software contributed to research, drafting, document analysis or advice.

Why human oversight becomes more important as AI improves

Greg Dickason, Executive Vice President and Chief Technology Officer at LexisNexis, made the architecture of trustworthy AI tangible during his keynote address at the AI Summit.

He began with an everyday experience.

His physiotherapist had asked whether she could use AI to record and produce notes from their session. He agreed. Within minutes of leaving, the resulting information had arrived. The speed was impressive. But that was not what stayed with him.

The question he found himself asking was: how much time had she spent actually reviewing those notes before they were sent?

That problem becomes more significant - not less - as AI gets better.

“I think we've reached a point in AI where we're starting to trust it maybe a little bit too much. Where we no longer, as humans, spend enough time in the loop. And the reason why that matters is that AI still gets it wrong - maybe only 1% of the time. But 1% in legal advice, in front of a client, in court, is not 1%.”

Dickason illustrated the point with a live demonstration: a small AI model trained on noughts and crosses. Without a harness around it, the model attempted an invalid move and the game failed. With a harness, the invalid action was caught before it could proceed. The harness did not play the game. It kept the model within the rules.

The analogy for legal AI is direct. A strong system is not simply a capable model; it includes the mechanisms that keep the model operating within appropriate boundaries.

The deeper argument was about where value actually lives. Models - even the best frontier models - are increasingly commodities. What cannot be replicated is 150 years of authoritative legal content and the legal intelligence layer that tells the model what matters.

“Authority is not a feature. It is the foundation of what we do.”

As models get better at absorbing more of the functions around them, the most important role of the harness becomes knowing when to get human attention - not constantly, not never, but at the moment legal judgment is actually required.

"Attention is all you need, not the model's attention, but yours. When it's appropriate for you to be in the loop."

From experimentation to practice: what scaling actually looks like

The customer panel that followed brought the theoretical architecture of trustworthy AI into contact with the reality of organisations trying to implement it.

A live poll of the room illustrated the gap. 43% said the biggest change they had seen over two years was the shift from experimentation to everyday use. Only 7% said the shift from individual adoption to firm-wide transformation.

Vanessa Bourne, General Counsel and Company Secretary at WaterNSW, described a three-part framework her organisation uses when evaluating new AI use cases, and when making the case to the board.

"It's the people, it's the governance guardrails, and the technology itself."

People: ensuring the team has the critical thinking to interrogate AI outputs, not just consume them. Governance: being explicit about which use cases are approved and which are ring-fenced. Technology: being equally explicit about which tools are authorised.

For WaterNSW's legal team, that currently means LexisNexis Protégé™ and Copilot - and nothing else.

The discipline behind that clarity came partly from experience. Like many organisations given access to Copilot, WaterNSW saw pockets of the business begin building their own agents independently, including the legal team itself.

"We were guilty of that ourselves. We set our own agent up where we pulled all previous legal advices into a ring-fenced SharePoint site. But the board and executive identified risks in that unfettered agent creation across the organisation. So, they've wound all of that back in."

Timothy Whincop, Director at Vocare Law, described how his firm's approach to governance had evolved from a compliance exercise into something more operational.

Two years earlier, the policy was simple: if you use AI, tell us, and only use this tool. That worked as a starting point. It did not scale.

"The governance becomes much more of an operational governance process and a continuous development process, as opposed to just a blanket set of rules."

The AI policy at Vocare is now reviewed every six months.

Timothy was also candid about the junior lawyer who was found using public Claude, convinced he wasn't putting confidential information in.

"Yes, but there's a bigger picture conversation we've had as a team about why we want this particular tool in place."

Associate Professor Wendy Bonython, Associate Dean of Learning and Teaching at Bond University, was designing curriculum for graduates who will practice in a world that looks nothing like today's.

"My crystal ball is proving remarkably unreliable looking that far out. So, the approach we're trying to take is, what are the principles that are going to get them through? What are the transferable skills?"

The students getting the best results from AI, she noted, are those engaging interactively - challenging responses, probing further, pressing for explanation. That iterative process, with all its friction, is where the learning actually happens.

The panel closed with a lightning round on assumptions.

What assumption about AI turned out to be completely wrong?

Vanessa: "That it was all about the technology and not the people."

Wendy: "That it would improve productivity in terms of quantity rather than quality."

Timothy: "That we'd be able to charge for it."

And what assumption had turned out to be right?

Timothy: "That it would improve to the point where you can trust source material."

Those answers, taken together, are probably the most efficient summary of where the profession currently stands, and what has changed.

The public sector dimension: governance before speed

The afternoon's Track 2 session illustrated the central challenge facing many organisations. When the audience was asked whether AI would significantly change how government legal and regulatory teams operate in the next three years, approximately 80% of hands went up. When asked who had a clear strategy and implementation roadmap to make that happen, the number fell to roughly 10%.

Lee Hickin, Executive Director of the Australian National AI Centre, was frank about what readiness actually requires.

"AI is not a thing that magically just solves a problem for you. It's really a skill that you have to learn — not in the sense that I've learned how to use Excel. It's a skill in the sense that I have to think about how I solve problems now in a world where I have mechanisms and tools that completely change that."

Dickason offered a longer historical view; the transition from steam-powered to electrically powered factories in the early twentieth century. Factory owners initially replaced steam engines with electric motors while keeping the vertical layouts built around a single power source.

The real transformation came only when they recognised that electricity made the entire layout negotiable.

"We're still thinking about AI revolutionising our current processes. We're still thinking in pre-AI terms."

Lee pushed back on the governance framing that dominates many regulated environments.

"I would much rather us be having conversations about practices - about changing the way we do things at a fundamental level - not adding in new governance layers. Half the time you look at what that is, it's just good business governance: understanding where your data comes from, understanding the risks, legal and otherwise."

"Transparency doesn't solve the problem. But it creates the space where you can then have a conversation about what is acceptable."

How AI is changing the economics of legal services

If Dickason's keynote examined the architecture of trusted AI, the afternoon discussion turned to what happens when AI-enabled processes begin operating at scale.

Robyna May, Chief Information Officer at McInnes Wilson Lawyers, described an experience familiar to many firms. Individual productivity gains are visible; faster summarisation, accelerated research, more efficient document extraction. But individual productivity is not the same as institutional transformation.

"What we're seeing is more of those component pieces giving us time savings. But it's a bit uneven because people's use of those productivity tools varies between individuals. What we want to get to is embedding those tools in shared workflows, so the knowledge is distributed rather than individual."

Mark Love, Legal Director at BAL Lawyers, emphasised another part of the value equation that can disappear when AI discussions focus too narrowly on speed: quality assurance.

“The proper use of AI gives you protection. It gives you the opportunity to look at particular issues more thoroughly, more deeply, and most importantly to test material before it goes out the door - to give you that warning that perhaps there is an issue you've missed. For small and mid-tier practitioners, that is an enormous benefit."

The word proper is doing substantial work in that sentence.

Indrajeet Aich, General Counsel at Challenger, brought the in-house perspective. When AI allows a legal team to produce a well-supported position on an unfamiliar question without instructing external counsel, the avoided cost can be quantified.

"We can actually now start to measure, in dollar terms, what we would have paid a panel firm for that document review or that research. That is part of how we justified the investment in tools like Protégé in the first place."

He was also candid about the in-house reality.

“Being in an in-house environment where you're at the coal face - speed, and I'm reticent to say this out loud, but sometimes the business wants speed over accuracy.”

Who captures the productivity dividend when AI reduces the time required to complete legal work?

"When computers made typewriters redundant, the productivity dividend was shared - in varying degrees - amongst all parties. I think we'll see the same thing here. It's too early to say exactly how. But a shared dividend, however it looks, is probably what history tells us to expect."

May added that the costs on the other side of the ledger are not stable either. Tools investment, learning time, and an impending shift from per-seat licences to consumption-based token charges.

"We do not want to be in a situation where we are training clients to expect a pricing model that we cannot sustain because the underlying cost model for AI is not yet settled."

What becomes scarce when legal production becomes cheaper

If AI makes certain forms of legal production faster and less expensive, what becomes scarce?

Not legal text. Not access to information. The more important scarcity is the ability to determine what matters.

O'Connor put the distinction clearly:

"If legal production is going to become less scarce, economic value will shift toward what cannot be automated. Judgment - knowing what matters, not just finding the information. Trust - knowing that an answer is sourced, explainable, and defensible. Context - the relationship with the client, the understanding of their industry and their appetite for risk. And accountability. Someone actually prepared to take professional responsibility for the outcome."

She added a fifth dimension: strategic imagination; using the capacity created by AI to pursue work that previously could not be done at all.

These characteristics have always mattered in legal practice. AI may make them more visible economically because it changes the cost of the surrounding production work. For generations, lawyers developed judgment by performing large volumes of it.

If AI absorbs a greater share of those tasks, the profession will need to think carefully about how expertise develops. A profession that values judgment still needs a way to produce people capable of exercising it.

What legal leaders should take from AI Summit 2026

Two hundred and sixty people attended AI Summit 2026. More than five hundred wanted to.

That gap is itself a useful data point. The legal profession across Australia, New Zealand and the wider Pacific is not engaging with artificial intelligence from a distance.

It is in the middle of working out what trusted, responsible adoption looks like - in firms, in-house, in government, and in the institutions training the next generation.

The AI Summit did not produce a single answer to every question, nor should it have. But a clearer standard is beginning to take shape.

The measure of successful legal AI will not simply be that it produces an answer quickly. The standard will increasingly be whether the result is authoritative, verifiable, appropriately governed and useful in the context in which a legal professional needs to act.

“The next phase of legal AI will not be defined by whether the technology works. It will be defined by whether the work it produces holds up.”

That is a considerably higher bar than adoption. It is also the bar legal practice requires.

From adoption to trusted legal AI

The central message from AI Summit 2026 was not that artificial intelligence has arrived in legal practice. The profession already knows that.

The more consequential question is what lawyers, and legal organisations build around it.

AI can accelerate research, drafting and review. Structured workflows can make successful methods repeatable. Technology can help legal teams test work, identify issues and create capacity for higher-value activity. But capability alone does not create trust.

LexisNexis Protégé™ is designed around that reality - combining AI capabilities with authoritative legal content and workflows built for the way legal professionals actually work.

Because the future of legal AI will not be determined by who can generate the most output. It will be determined by who can turn AI capability into legal work that professionals can verify, defend and trust.

See how Lexis+ with Protégé can support trusted, AI-enabled legal workflows in your practice.