A year ago, the most sophisticated AI use around ASX announcements with JORC content was often limited to spell-checking or smoothing awkward phrasing. Today, Competent Persons (CPs), lawyers, and company secretaries are increasingly reviewing AI-generated drafts and AI-flagged inconsistencies as part of live disclosure workflows. The pace of that shift is unusual. The legal framework around mineral asset disclosure has not moved with it. The JORC Code's first major revision since 2012 continues to progress, but the technological environment it will eventually land in is materially different from the one that existed at the time of the Code's last update.
That speed has not gone unnoticed at the top of the legal system. In May 2026, Chief Justice Andrew Bell of the Supreme Court of New South Wales delivered a speech on the potential impact of AI on judicial independence and the rule of law. He addressed hallucinations, automation bias, the risk of de-skilling, and what he described as "techno-capture" by the private actors who build and control the underlying models. His point was not to chase a trend. It was to identify structural limits on the use of AI in legal decision-making. The law has not formally changed. But the way the legal system is thinking about AI is changing quickly and materially.
For mining and resource entities operating inside one of Australia's most complex disclosure regimes, the question is now sharper than it was a year ago: not whether to think about AI, but how to use it without falling on the wrong side of the regulator, the law, or the standards the profession sets for itself.
The two functions of JORC
For those unfamiliar, the JORC Code performs two distinct regulatory functions. First, it sets the professional competency benchmarks and prerequisites for interpreting and estimating mineral assets. Second, it provides the disclosure pathway through which technical information reaches investors and capital markets, intersecting with the ASX Listing Rules, ASIC guidance, and the Corporations Act. Where AI can usefully contribute differs sharply between the two.
Competency is the wrong place to look
Competency under JORC does more than filter the quality of reporting. It also performs a legal function. The Corporations Act requires reasonable grounds for forward-looking statements, and in a sector defined by speculative assets, long development cycles, capital intensity, and embedded uncertainty, the competency of the CP often becomes a primary anchor for the reasonableness of technical assumptions. In practice, the experience and reputation of the CP making the statement can matter as much as the statement itself.
AI cannot demonstrate years of experience, hold professional membership, or exercise judgement on incomplete data. And, most importantly, it cannot be held responsible.
That makes the limits of AI on the competency side of JORC relatively clear. AI cannot demonstrate years of relevant experience. It cannot hold membership in a recognised professional body. It cannot exercise judgement under conditions of incomplete data. And perhaps most importantly, it cannot be held responsible. Whether AI can support a CP's technical work is a separate question. It cannot replace the accountability and legal foothold that the CP provides.
The more useful question is whether AI has a meaningful role in the disclosure pathway that JORC requires technical information to travel through.
The pathway problem
A technically sound and complete JORC report is not, of itself, a compliant market disclosure. The information must also satisfy the ASX Listing Rules, ASIC guidance including Information Sheet 214, and the disclosure tests in the Corporations Act. None of those frameworks were drafted in the language of geology or engineering.
This is the structural awkwardness of JORC reporting. The CP is the technical author of a document that sits inside a legal framework designed for, and administered by, lawyers and regulators. JORC regulates the manner of technical reporting. The moment that report becomes part of an entity's public disclosure, a different set of obligations attaches.
The practical consequence is familiar to anyone who has worked on an announcement that includes a JORC statement. A draft prepared by the CP passes through management, the board, legal, and the company secretary before it is released. The CP, responsible for the technical elements, is asked to consent to the final version as a fair and accurate reflection of their work, even where the surrounding announcement has added context, timing, or commercial framing they did not author.
The friction is structural, not procedural, and it is where many disclosure errors arise. It is also where AI can deliver its most significant benefits.
The regulators are already using it
CPs and advisers should be in no doubt that the technology has already arrived on the regulator's side of the desk. ASIC has published an AI Transparency Statement confirming the development and deployment of AI across its regulatory activities, while noting that its use in surveillance and enforcement may not always be disclosed.
The change is not just one of pace. The cost of detecting disclosure defects is falling sharply, and what is detectable is expanding with it. Longitudinal comparison across years of announcements, event-driven ramping conduct, pattern detection across peer issuers, and real-time cross-checking of JORC content against market commentary and forward-looking statements are now technically and economically feasible for regulators. The defects that used to get caught are not the only ones that can be caught now.
Mineral asset disclosure already operates inside one of Australia's most complex disclosure regimes. It has now become materially easier for regulators and compliance teams to interrogate it at scale.
Three choices, but only one clear path
This leaves resource entities and their advisers in a difficult position. Regulators are building stronger AI-enabled compliance capability, the legal system is openly examining the risks of AI dependence, and boards still need to get announcements out accurately and on time. There are three options, and only one is defensible.
The first is not to use AI at all. The CP, the lawyers, and the company secretary keep doing the work the way they always have. Drafting cycles remain slow. Memory-based and manual reviews remain the test for consistency against years of prior announcements and against peer disclosures. Every lodgement is made while hoping the regulator's models do not surface something the manual review missed. The asymmetry is no longer theoretical. It is the operating reality.
The second is generic AI integration. A general-purpose LLM is brought into the workflow to draft, review, or compare, with manual oversight asked to catch what matters. This is the territory Chief Justice Bell's address is directly concerned with. Hallucinations are real and recurring. Automation bias is well documented. The legal profession is now treating uncritical reliance on AI outputs as a professional risk in its own right. Undertaking the diligence needed to have assurance that general-purpose models are reliable is likely to wipe out any productivity gains they were brought in to deliver.
The third option is to use purpose-built AI. A general-purpose LLM is an engine, not a finished vehicle. If it is to be useful in this environment, it needs domain-specific regulatory context, controlled knowledge inputs, role-sensitive review pathways, and human checkpoints that preserve responsibility where the law expects it to sit. Used that way, AI can help review and generate disclosure that understands JORC as it actually operates: not as a standalone code, but as part of a broader disclosure architecture.
This is the standard the regulators are building on their side, and a capability that entities with disclosure obligations should expect on theirs.
Purpose-built AI is harder than it sounds
Purpose-built AI is the defensible option, but it is genuinely difficult to build. JORC disclosures sit across multiple domains that are usually handled by different people and different tools. Technical content is the work of geologists and engineers. Disclosure obligations and governance sit with lawyers, directors, and company secretaries. The regime demands both, applied together, against a moving body of regulator commentary and enforcement posture.
A system that handles only part of that landscape is not a solution. Nor is a general-purpose model with a JORC-shaped prompt. The hard part is not the model itself. It is encoding the regime, keeping it current, and directing the right issue to the right framework so the system can assist a CP, a company secretary, and legal counsel within the same announcement.
That requires something closer to regulatory practice than generic software deployment. The value is in understanding how the regimes connect and how the regulator actually reads what entities say. Without that, an AI model produces output that looks competent on individual obligations and is structurally unfit for the purpose.
Is there a solution?
RGC Advisory's team includes former ASIC mining and resources specialists with oversight experience across thousands of disclosures under the JORC Code, the ASX Listing Rules, and the Corporations Act. That regulatory lens sits with a relatively small group of legal practitioners in Australia.
The pattern is consistent. It is rarely the geology, engineering, or completeness of a JORC statement that creates the problem. It is the gap between a technically sound JORC statement and a legally compliant market announcement. That is where news is delayed, transactions are queried, and disclosures can stall.
No generic AI model closes that gap on its own. But a well-built one can make the path across it shorter, better evidenced, and less dependent on the CP being asked to operate as a quasi-lawyer.
That is what we built RGC Onside for: a purpose-built AI designed to help ASX-listed entities navigate JORC, the Listing Rules, and the Corporations Act together, while building the contemporaneous evidence base the regulator may one day ask to see.
Try it for yourself. Upload an announcement and see what it finds. If we can find the issues, the regulator probably can too.
