Who Owns What the Machines Learn From Us?

One new thing is being introduced into our health systems. It will change everything it touches — and the most valuable thing it touches has no owner.

I want to walk you through a pipeline. Five stages, each one perfectly legal, and at the end of it the accumulated clinical knowledge of an entire country sits on the balance sheet of a foreign company. No stage requires anyone to do anything wrong. That is what makes it worth your attention.

You may not work in the health sector but if yours relies on middleware to operate then the same type of harvested digital assets scenario is fast heading your way.

Stage one: the seams disappear.

For thirty years, health systems have run on middleware — the unglamorous connective tissue of integration engines, transcription services, order interfaces and results routing. Clinicians curse it daily, and fairly. But every layer was also a seam: a contract boundary, an independent log, a place where a data flow could be inspected by someone other than the party who created it. The seams were the system’s independent witnesses.

A new generation of agentic AI platforms — led by one of the world’s largest technology companies, whose AI-first clinical system is already live in North America and the UK — removes them. The AI listens to the consultation, writes the note, drafts the orders, routes them and feeds the billing engine. One vendor, one cloud, one semantic layer, where a dozen contracted interfaces used to be. It genuinely reduces clinician burden. It also means that when something goes wrong, every piece of evidence sits inside the vendor’s stack — while the legal duty sits with the doctor who approved the AI’s draft, one of forty in that shift’s review queue.

Stage two: consent becomes a legacy fiction.

The consent you signed on admission was drafted for an electronic records system. It says nothing about continuous ambient AI listening to your consultation — including the things you say beyond clinical relevance, and the family member beside you who consented to nothing. Australia’s informed consent standard asks what a reasonable patient would consider material. Would you consider it material that an AI drafted your medication order? So would I. Nothing in current deployment tells you.

Stage three: your data exits the law through a one-way gate.

Here is the wicked one. Europe treats identifiability as a spectrum, assessed against what re-identification technology can do now and next. Australia treats it as a binary, assessed once: the moment data is classified as “de-identified,” it exits the Privacy Act entirely — no consent, no cross-border limits, forever. Our regulator recently examined a case where de-identified patient imaging was shared, without patient consent or notice, to train a commercial AI model. No action was taken. The data was no longer “personal information.” The exit was clean.

The trouble is that the classification is made once, at a point in time, while AI drives the cost of re-identification down a curve, forever. AI sits on both sides of the gate — the technology making re-identification cheap, and the biggest consumer of the data walking through it. We decide protection at the moment of maximum ignorance, and the decision cannot be revisited.

Stage four: the residue becomes Knowledge — and Knowledge belongs to no one.

My first job, in 1971, was at the CWS, where I watched ledgers give way to dividend stamps. The lesson has never changed: the value was never in the transactions — it was in who held the recording layer. Middleware vendors learned it a generation ago: you never needed to keep the raw data if you kept the metadata. The metadata was the gold. Agentic AI completes the arc: it needs neither. What it keeps is derived structure — the weights, the patterns, the clinical intelligence distilled from watching a whole health system in motion. Call it Knowledge.

Now watch each body of law wave it through. Copyright needs a copy — courts have accepted a trained model doesn’t store copies of what it saw. Privacy needs identifiability — severed at stage three. Confidence needs a confiding relationship — dissolved by contract. And theft needs a thing — and Australian law has said for eighty years that information is not property. The High Court held in 1996 that you have no property right even in your own medical records. The extraction commits no wrong, because every wrong requires a category the extraction has been engineered out of.

Stage five: title crystallises offshore — and becomes treasury.

In 2025 the United Kingdom passed the Property (Digital Assets etc) Act, opening a “third category” of personal property for digital things that are neither possessions nor contractual rights. It was written for crypto-assets. Statutes go where litigants take them. A well-constructed knowledge artefact, engineered to the Act’s criteria, can be brought before English courts and come out the other side as property — good against the world, transferable, securitisable.

And the machinery to monetise it is already running. BlackRock’s Larry Fink has put tokenisation at the centre of the firm’s outlook — “roughly where the internet was in 1996,” the update to “the plumbing of the financial system,” the beginning of “the tokenisation of all assets.” SharpLink — the multi-billion-dollar Ethereum treasury company chaired by Ethereum co-founder Joseph Lubin, with a former BlackRock executive as CEO — has published the corporate playbook: hold the digital asset not as an investment but as productive financial infrastructure, deployed, collateralised, put to work. To be clear: their treasuries hold ether, not health data. But the playbook is public, and the extrapolation writes itself. Whatever the law recognises as property, this machinery can tokenise. Whatever is tokenised becomes balance-sheet material — an appreciating reserve, priced continuously by AI agents paying to query the intelligence it encodes.

So: the aggregate clinical knowledge of the Australian population, derived from our bodies, our consultations, our outcomes — held as the treasury asset of a foreign entity, under the law of a third country.

Who owns this, in the new world order? At the moment, the honest answer is: whoever constructs the claim first.

The opportunity hiding inside the threat

I’m not writing a doom piece. The same five stages, read in reverse, describe the opportunity — and the question that decides who takes it is the one I’ve asked in every room for thirty years: what would make it worth it? For each party, there is now an answer.

For corporates — including the vendors themselves. Trustworthiness requires truthfulness, and in this market truthfulness has an architecture: independent consent and audit interfaces wherever a contractual boundary used to exist — seams by design, not by accident. That, plus a licensed, royalty-bearing position in the Knowledge you help create, is not a compliance cost; it is the trust product that wins sovereign procurement in every health system on earth. The vendor suspected of quietly acquiring title eventually wins none of them. The window to choose is now, before mandatory standards choose for you.

For governments. Australia got minerals right: the Crown owns what is in the ground, extraction happens under licence, and royalties flow. For knowledge derived from public systems there is no Crown title, no licence, no royalty — because the law says there is nothing there to own, right up until a foreign court says there is. The fix costs nothing but drafting: procurement clauses vesting title to derived models and analytics in the Crown, situs rules keeping Australian-derived knowledge under Australian law, a resource-rent frame for the asset our health system generates every day. Then tokenisation works for the public — a sovereign knowledge treasury, not a foreign one.

For the humans who generate the knowledge — every clinician, and every health consumer. This is the part I care most about. The knowledge in question does not come from machines. It comes from a nurse’s forty years of pattern recognition, a doctor’s judgement call at 3 a.m., and from you — your body, your history, your consultation, your outcome. AI offers probabilistic approximations to fill deterministic gaps; the one thing it cannot supply is the human who determines what is actually real. That judgement is your point of difference, and it can be measured, verifiable and validated, and captured as an asset owned by the person who made it — tokenised, provenance-bound, inalienable at its root. What we need is a new species of right the digital economy does not yet have: a personal copyright in validated experience. The UK’s new property category cuts both ways — the same statute that lets a corporation claim the residue lets a citizen-side structure crystallise ownership first. First construction wins. I would rather it be ours.

One new thing changes everything. The only durable response is to be able to see everything it changes — and to make sure the humans the value comes from hold title to their share of it.

This post distils a longer discussion paper on AI, consent, property and the Australian health system in motion. If you want the full argument — the case law, the statutes, the counter-moves — reply or message me and I’ll share it. Nothing here is legal or financial advice; the named companies’ strategies are cited from their public statements, and no intention regarding health data is attributed to any of them.