“For thirty years, the internet economy trusted a single act, the click, as its proof that a real person was there. Artificial intelligence has made that proof free to counterfeit. We think what replaces it is the oldest instrument in commerce, rebuilt for the age of AI: a way to prove a real human action, then value it and settle it.”
When a click was proof
For thirty years, a click was proof enough of a person online.
Every trade between strangers rests on one hard question: is there really someone on the other side, and are they who they say they are? For most of history, commerce answered that question with an instrument: a seal, a signature, a stamped letter. The internet economy answered it with a click. When someone clicked, the economy assumed a real person was there who wanted something and might pay, and it went on to build almost everything on top of that one assumption.
That assumption ran deeper than most people ever noticed. The price of an advertisement, the ranking of a search result, the score a fraud team assigns, the decision to ship an order: follow any of them down and you reach the same quiet belief, that behind this click stands a human being.
The click worked as proof for one reason, and it is worth saying plainly. Faking a person used to be expensive. To manufacture clicks that passed for human, you needed real-looking accounts, real-looking devices, and real-looking behaviour, and all of that cost real money. The expense itself was the security, even though almost nobody ever wrote it down.
AI made fakes free
But AI made a fake person essentially free.
Now, with a modern model, a click, a session, a review, a face, or a voice can be produced by software, at any scale, for almost nothing. The account looks real, the history looks plausible, the face looks human, and yet there is no person there at all.
Today, bots generate the majority of all the traffic on the web, and a single click can no longer tell you whether a person or a script is behind it.
The usual word for this is fraud, and that word is far too small. Fraud is a handful of bad actors slipping past a system that still works for everyone else. What has happened here is different in kind, because the system's basic unit of measurement, a click that stood for a person, has quietly stopped meaning anything, and no filter can restore a meaning that is simply gone.
What breaks next
AI-driven fakes break the ad market, then commerce, then settlement itself.
The click held up far more than advertising. It quietly propped up three different parts of the internet economy at once: the price of an ad, the trust in a review, and the winning of a customer. A fourth part, machines paying machines, is only now being built, and it needs the very thing the click can no longer supply. Take the click away, and all of it comes loose together.
Advertising, reputation, and customer acquisition all leaned on the click to know a real person was there. Machine-to-machine settlement is the new floor being added, and it needs the same thing the click can no longer give.
Advertising breaks first, because advertising is where the click was money. Businesses spend about $1.1 trillion a year on advertising, and roughly three-quarters of that money is priced by the click, the view, and the install. A price needs a real unit to count. Once the click stops standing for a person, the buyer keeps paying by the unit while the unit itself hollows out.
Reputation breaks next. A review, a rating, and a follower count were trusted because they were once cheap to earn and expensive to fake. AI has made them cheap to fake as well, so a five-star average now tells a shopper almost nothing, and the ranking that used to sort the good from the bad slides toward a coin flip.
Customer acquisition breaks after that. To sell to someone, a business first has to reach a real person who genuinely wants what it offers. When both the account and the intent behind it can be manufactured, a growth team ends up paying for sign-ups that were never people, with no way to tell the real ones from the rest. It overspends for a while, and then it pulls back.
Settlement is the last to break, and it is the one that has barely begun. Software has started to buy from software itself: agents that book, subscribe, and pay on a person's behalf, at machine speed, with no human checking each transaction, and analysts already put this coming machine-to-machine economy in the trillions. Here the danger cuts both ways. A buying agent cannot tell whether a real person or business truly stands behind an offer, and a selling agent cannot tell whether it is dealing with a genuine customer or a malicious script. Because neither side can prove a human, neither side can safely close the transaction, and a market that could be vast cannot form at all.
All four failures come from a single hole, showing up in four different places. Advertising, reputation, acquisition, and settlement rested on the same quiet fact, that a real person was on the other end, and AI knocked that fact out from under all of them at once. The ground keeps shifting, too, because every new model makes a fake a little cheaper and a filter a little weaker.
What AI cannot make
The thing that replaces the click has to be something AI cannot make.
Whatever takes the click's place has to be something AI cannot manufacture, and that rules out almost everything we might reach for first. A face can be forged, a password stolen, a login automated. A bot filter tries to catch fakes after the fact, but an AI fake now behaves exactly like a real user, so there is nothing left to catch. A captcha or a second code only proves that someone can pass a test in the moment, not that a real person with a real history stands behind the account. And a platform that vouches for its own users is both referee and player, because it earns more when its user numbers look bigger, so it has every reason to wave a doubtful account through. No matter what wall we build, the next model climbs it.
One thing, though, does not fall to a bigger model, and grows more valuable the more capable AI becomes: a real life, lived over real time. Years of genuine activity, spread across real accounts and tied to a real history, cannot be produced on demand, because producing them would mean having actually lived them. It is the same reason the art market is worth tens of billions of dollars. A perfect copy of a painting is worth almost nothing, while the original, the one whose history can be traced, is worth a fortune. What is scarce is not the image but the provenance, and a person's real history is exactly that: provenance, for a human being.
For most of the internet's life, proving you were a real person was expensive and clumsy, which is why almost nobody bothered. Two industries changed that, racing in parallel and crossing at nearly the same moment. Artificial intelligence drove the cost of faking a person down toward zero. At the very same time, and quite separately, cryptography drove the cost of proving one down toward zero as well. The tools we needed had been sitting under our nose the whole time.
Faking a person got cheap years before proving one did.
Faking a person got cheap first, the moment AI arrived. Proving one caught up only later, as cryptography matured. The gap between them, when fakes were cheap but proof was not, is the dangerous window we are now closing.
For thirty years, browsers have wrapped their traffic in an encrypted session called TLS, the small padlock beside the web address, so that a bank or a shop could be certain who it was talking to. That session leaves behind a cryptographic record that cannot be altered after the fact. A newer branch of the same mathematics, the zero-knowledge proof, lets a person lift a single fact out of that record, that this account is more than ten years old, or that this balance once cleared, and prove it to anyone while revealing nothing else. A proof built that way is what we call a sealed proof. The fact itself travels and can be checked in seconds, while the data behind it never leaves your device. The only thing that survives a smarter machine is a proof it cannot make, checked by someone with nothing to gain from the answer.
The missing instrument
Every market in history had to invent a way to trust a stranger.
This problem is not new at all, but the oldest one in commerce wearing new clothes. Every market that ever grew beyond a single village ran into the same wall: how do you do business with a stranger you cannot see, and may never meet again? For three thousand years, the answer was an instrument, a small and hard-to-copy device that let one person trust another without knowing them.
Every market that grew beyond a village invented an instrument to trust a stranger. The internet economy built TLS, which proved the connection was private, then skipped the one instrument that proves a person is really there.
A merchant pressed a wax seal that only his own ring could make. A signature put a person's own hand on a promise. A bill of exchange let a trader in one city be paid in another. Double-entry books gave both sides of a deal a record they could check against each other. The clearing house placed a trusted party between buyer and seller, so that a trade could settle even when the two did not trust each other directly. Each of these did the same quiet work for its age: it let value move safely between strangers.
The internet economy built one such instrument, and then it stopped. In 1994, that small padlock in the browser, TLS, made it safe to send a card number to a stranger's website, because it proved the connection was private and the site was really itself. Online commerce did not take off and then bolt on the padlock. It waited for the padlock, and it only grew once the padlock had arrived.
The padlock every browser shows: proof, for thirty years, that a connection is private, though never that a real person sits at the other end.
The padlock proved the pipe between two computers was private. It never proved a real person sat at the end of it. The click stood in for that missing proof, and it held only while faking a person stayed expensive. That expense is gone, so the instrument every market eventually builds is finally due: a way to prove a real, valuable human is on the other side of a trade, and to settle what that human is worth to the business reaching them.
The same mathematics behind the padlock now supplies it. HTTPS secured the connection. A newer layer, the zero-knowledge proof, secures the person, and it rides on the very session HTTPS already protects. The proof stays open and checkable by anyone, the way the padlock's mathematics belonged to no one. What is left is work, not trust. Someone has to run the software that checks each proof, matches a real person to the business that wants to reach them, and moves the payment once the action is verified. We run that marketplace, we do the matching, the attribution, and the settlement, and we charge for the work. We never charge for the trust, because the trust is now mathematics, and mathematics is cheap.
What moves between the two sides is easy to name: a real person, with the traits they claim, taking an action a business will pay for. The proof is what makes that action countable, so a business can fund an offer, a real person can complete it, the proof can be checked, and the payment can settle against an action that truly happened. For the first time the internet economy can price a human action by whether a human really took it.
This is a different kind of value from the one the internet economy already trades. It has always guessed at people, building a profile out of what they click and browse, a model of what someone might do next. A sealed proof works the other way. It records what a person actually did, checked and true, and what someone truly did is worth more than any prediction of it. That record is the one asset a smarter machine makes more valuable rather than less, because the better a model gets at imitating people, the more a proof of a real action is worth.
How early this is
We are already carrying it, at the very start of the curve.
We would rather show this than promise it. We have been building the rail for a while now, it already carries real weight, and we are the first to admit how early it still is.
Small today, and that is the point: a better instrument almost always starts small. The padlock secured a handful of checkout pages long before it wrapped the whole web.
Three hundred and twenty thousand people already carry sealed proofs of their real lives, businesses already pay to reach them, and more than a hundred thousand dollars in verified transactions changes hands every week on the rail. Set against the $1.1 trillion that still flows every year through clicks that no longer mean a person, those figures are tiny, and that is exactly what we would expect, because a better instrument almost always starts small before it spreads.
For thirty years, the internet economy learned everything it could about people and paid them back in free services rather than in money. It could afford to do that because it could not tell a real person from a convincing fake, so to the machine a genuine human and a bot were worth the same amount, which was essentially nothing, one at a time. We can finally tell them apart. Once a business pays to reach and reward a person it can prove is real, a real life becomes the scarce thing worth reaching, and a proven action becomes the unit it pays for, so for the first time the person is paid for their own life online instead of giving it away for free.
For the last two years I helped build the T‑REX community, hundreds of thousands of people proving small facts about their online lives, and for most of that time the hardest part of the job was explaining why it mattered. Then agents arrived. I watched my own agent draft my messages, plan my trips, and negotiate on my behalf, and I watched it get better every month. What it could not do, and will never do, is spend years earning the history it was spending in my name. My agent can act for me. It cannot have been me. Somewhere in those months, the explanation I had struggled with became obvious: when software can do everything a person does, the only thing left worth proving is the person.
AI can do anything you do. But AI cannot be you. That has always been true, and for thirty years it did not matter, because the internet economy had no way to charge for the difference. It can now. The one thing a machine cannot fake, a real human life built over real years, turns out to be the one thing most worth proving, and now the one thing worth paying a real person for. Building the instrument that proves it is the work we have given ourselves, and we think it is the most important work on the post‑AI internet.