Provenance
When every credential can be forged for free, the only thing left to sell is your name on the line.
Hey. My random inspiration is the main driver for this edition.
Last month, a short story called “The Serpent in the Grove” beat 7,806 other entries to win the Caribbean region of the Commonwealth Short Story Prize.
The judges, chaired by a real novelist, praised its “precise yet richly evocative” language. Then a Wharton professor ran it through an AI detector, and it came back flagged at 100%. Three of the five regional winners lit up the same way.
Here is the part almost everyone scrolled past. In June, the Commonwealth Foundation reviewed the allegations and stood by all of its winners.
Sit with that for a second. Faced with the one question a literary prize exists to answer (did a human make this?), the institution whose entire job is to answer it chose not to. It could not. So it defended the certificate instead.
The internet read this as a story about AI learning to write. It is not. It is a story about what we were grading the whole time.
The prize was never measuring the writing. It was measuring a proxy. And the proxy just stopped working.
What We Were Actually Grading
Here is the thing nobody wants to say out loud.
A literary prize cannot see whether a human had something to say. That is invisible. It is locked inside a skull. So the prize measures the nearest visible thing instead: does this read like the kind of writing that wins prizes? The judges were not duped by a robot in a trench coat. They did exactly what judges always do, which is grade the surface and trust that the surface points at a soul underneath.
For a few thousand years that trust was safe, because faking the surface was expensive. You could not produce prize-shaped prose without the years of reading and failing that tend to come bundled with actually having something to say. The surface was a reliable stand-in for the substance.
So we built the entire civilization on stand-ins.
The diploma is a stand-in for “this person can think.” The byline is a stand-in for “this was worth publishing.” The portfolio stands in for “they can really do the work.” The blue tick, the follower count, the “as seen in,” the ten years of experience, the glowing case study, the prize. Every one of them is a cheap signal we accept in place of an expensive truth we cannot afford to check directly.
And every one of them works on exactly one condition.
A proxy is only worth what it costs to fake.
The Proxy Collapse
This was the month AI shipped more frontier models than any month in history. I lost count somewhere around the sixth launch. And the surface they produce is no longer “good for a machine.” It is just good.
Look at what that did to the cost of faking. Ahrefs studied close to a million new web pages and found roughly three-quarters carried detectable AI writing. Graphite, an SEO firm, combed through more than 60,000 articles and found that by late 2024, over half of all newly published English-language articles were primarily AI-written. Europol has warned that as much as ninety percent of online content could be synthetic by the time you read this.
Competent prose, free. Competent code, free. Competent design, competent strategy decks, competent cover letters, all of it now infinite, instant, and good enough to beat eight thousand humans for a literary prize.
The cost of faking the surface just fell to roughly zero. Which means every proxy resting on “this looks like the real thing” is collapsing at the same moment. I call it the Proxy Collapse, and the literary prize is simply the first one naive enough to publish the lab results.
Your hiring funnel is running the identical scandal right now. It just has not been handed to a journalist yet.
Your Hiring Funnel Is Running the Same Scandal
Watch the numbers, because they are louder than any take.
LinkedIn now clears around eleven thousand job applications every minute, up forty-five percent in a year, almost entirely because generative tools let candidates fire off tailored applications at machine speed. The average opening draws 242 applications and gives any single applicant about a 0.4% chance. 90% of HR managers say their workload has increased due to the flood. Most say hiring now takes longer, because a stack of “perfectly matched” applications forces them to do by hand the verification the resume used to do for them.
The resume was a proxy. It died quietly, in the same quarter as the literary prize, and nobody held a funeral.
Then it got worse, because the forgery climbed one level up. It is no longer just the work that is faked. It is the worker. Palo Alto’s researchers found that someone with zero image-editing skill can stand up a fake job candidate, face and voice and CV, capable of passing a video interview, in about seventy minutes. A GetReal Security survey found 41% of risk and security leaders admit their company has already hired and onboarded a fraudulent candidate. Gartner expects that by 2028, one in four candidate profiles worldwide will be fake.
So Google and McKinsey did the thing that looks like a step backward and is actually the only sane move. They brought back the mandatory in-person interview.
When the proxy can be forged for free, you stop trusting the document and start verifying the person.
Why a Better Detector Won’t Save You
The reflex, of course, is to build a better detector. Granta’s own publisher reached for an AI to catch the AI: she fed the suspect story to Claude, which came back saying it was almost certainly not the unaided work of a human. Read that again slowly. We have reached the point where a human being needs a machine to tell him whether another machine wrote something.
It does not even work. The detectors are a coin-flip wearing a lab coat. A 2026 study out of Sultan Qaboos University reported that leading tools achieved 61% to 69% accuracy and found their performance on blended human-and-AI text collapsed to nearly zero. False-positive rates run from around 2% on the best tool to nearly 15% on the worst. And they punish the wrong people: writers working in a second language get falsely flagged two to three times as often, because careful, polished, slightly formal prose is exactly what the machine learned to imitate.
Which means Jamir Nazir’s defence, that detectors are unreliable and flag honest writers, is correct. And that is not the comforting part of the story. That is the terrifying part. It is no longer possible to verify the proxy in either direction. You cannot prove the prose is fake. You cannot prove it is real. The signal carries no information at all. The Foundation did not stand by its winners out of conviction. It stood by them because conviction is no longer available to anyone.
This is the trap. The harder you chase a detector, the more obvious it becomes that detection is a losing game. The thing that survives the Proxy Collapse is not a sharper detector. It is the opposite of one.
What the Art World Learned, the Hard Way
The art world worked this out a century ago, and it cost them eighty million dollars in tuition.
The Knoedler gallery was the oldest in New York. Over fourteen years it sold roughly sixty newly discovered masterpieces, Rothkos, Pollocks, Motherwells. Every single one was painted in a garage in Queens by one immigrant artist working from a photo book. Credentialed experts vouched for them. The gallery’s own president believed in them to the end. The brushwork was good enough to fool the people whose entire profession is not being fooled.
Here is the line worth carving above your desk. A Pollock painted in that garage is worth nothing. The identical canvas, same paint, same gesture, with a documented chain of custody and an expert who stakes their name on where it came from, is worth millions. The atoms are the same. The brushstrokes are the same.
The only difference is provenance: the verifiable story of who made the thing, and who is willing to be ruined if that story is a lie.
When copies are free, the original is not valuable because it looks better. It looks identical. It is valuable because someone can prove where it came from.
The Provenance Premium
Value did not disappear in the Proxy Collapse. It moved. It moved off the artifact and onto the provenance. Off the thing and onto the vouch.
And the market is already repricing it in real time. Remember that statistic hiding in the hiring data. More than seventy percent of recruiting leaders are now conducting interviews in person for the express purpose of confirming that the human is who the AI-polished application claims. That is not nostalgia. That is the Provenance Premium showing up as a line item. They are paying, in time and money, for the one thing a document can no longer give them: proof of origin.
Stop Competing on Output
So stop asking how to compete with infinite free output. You cannot. Not today, not ever. Start asking how to build provenance.
Whatever you make, stop competing on the artifact. You will lose to infinity. Anyone can generate a brilliant paragraph, a clean deck, a flawless function this afternoon. What no machine can hand you is a public, time-stamped body of work where people watched you be right, change your mind, ship, and take the hit when you were wrong. Compete on the track record, not the output. Almost nobody can show ten years of being publicly accountable for what they said.
And whatever you are buying, hiring, or reading, flip your default. The polished surface now tells you nothing, because the polish is free. The resume, the portfolio, the writing sample, the finished thing: all of it is just the artifact, and the artifact has stopped carrying information. So stop grading the finish. Ask for the chain of custody instead. What did this person actually decide, what happened next, and who will put their own reputation on the line to vouch for it. Treat the flawless artifact as forged until a person you trust has staked something real on it. Your trust has to migrate from the object to whoever will be embarrassed if the object is a lie.
Which leaves the one move that matters most, and it is not producing. It is vouching. The scarce, valuable, un-automatable act is being the one who says “I read all 7,806, and I am putting my name on this one,” and being right often enough that your name starts to carry weight. A model cannot do this. Not because it lacks taste, but because it has no name to stake and nothing to lose. It cannot be embarrassed. It cannot be ruined. Skin in the game is the one input AI is structurally incapable of having.
When the work is free, the only thing left to sell is your name on it.
I argued in AI Generalist that the edge was connecting things AI cannot connect. This is the harder-edged sequel: the edge is vouching for things AI cannot vouch for, because it has no reputation to burn.
Six Moves for This Week
That is the worldview. Here is the Monday-morning version. Six moves to start building your own provenance this week.
Timestamp everything. A dated, public trail of your work and your calls is the cheapest provenance there is. The point is not the post. The point is the date on it.
Show the chain of custody, not just the output. Publish the decision, the draft, the thing you got wrong, the why. Anyone can ship a clean artifact now. The mess is the part that proves a human was there.
Collect vouchers. Provenance is social. Line up the named people who will stake their own reputation on you, and be that person for others. A vouch from someone with something to lose is worth more than any certificate.
Make claims that can be checked. Real numbers, real outcomes, things a skeptic could verify and you would be humiliated to have faked. Verifiability is the new credential.
Be wrong in public, on the record. The single most un-fakeable signal you own is a track record of accountable mistakes. A machine will never do this. Use the moat while it is wide.
Own a verified identity and channel. In a world heading for one fake profile in four, a durable, provable “this is really me” is not vanity. It is infrastructure.
If you do nothing else, do the first one this week. Provenance compounds, and it only starts counting from the day you begin.
The Certificate Is Already Gone
Back to “The Serpent in the Grove.” Maybe Jamir Nazir wrote every word from his childhood in rural Trinidad, exactly as he says. Maybe he did not. The point that should keep you up is that you genuinely cannot tell, the Foundation cannot tell, the detector cannot tell, and the not-being-able-to-tell is the entire event.
Because the prize was meant to stand in for “a real person had something real to say.” The moment that stand-in is in doubt for one winner, it is in doubt for all five, including the ones who bled for it. The forger does not just steal a prize. He bankrupts the certificate for everyone still holding one.
When the Commonwealth Foundation chose to uphold its winners, it was not defending Nazir. It was defending the idea that the certificate still means something. Quietly, that idea is already gone.
The future does not belong to the best generator. Generation is free. It belongs to whoever can still say “I checked this, it is real, and my name is on it,” and have that sentence carry weight.
Provenance is the last scarce thing. Start building yours now, while everyone else is still arguing about whether the robot can write.
Thanks for being here. Thanks for reading, for the replies, for staking a little of your own name every time you forward one of these. That is the whole game.
Post-Credit Scene
My fair objection: provenance can be faked, too. The Knoedler forgeries came with forged provenance, an invented story about a mysterious collector, and it held for fourteen years.
So provenance is not unforgeable. What broke it was not a better document. It was consequences: the lawsuits, the closure, the experts whose names were ruined. That is the real engine. Provenance is the last scarce thing not because it cannot be copied, but because it is the last thing that still comes with a bill when you lie.
Five things worth your attention this week, all circling the same drain: what is real, who decides, and why the expert is so often the easiest one to fool.
Essay
Homo Agenticus by Rohit Krishnan (Strange Loop Canon, May 2026). On living alongside AI agents as a new species in our economic life. His quieter point is the one that matters here: automate one step and the bottleneck does not vanish, it moves to the next step. Right now it is moving from making to verifying. Go stand where it lands.
Documentary
Made You Look: A True Story About Fake Art (Netflix). The Knoedler story in full. Watch credentialed experts vouch, with total sincerity, for masterpieces painted in a Queens garage. This is your 2026 hiring pipeline with better lighting and a sadder ending.
Book
Caveat Emptor by Ken Perenyi. The memoir of an American art forger who fooled Sotheby’s and Christie’s for decades and never got caught. He explains, with receipts, how the surface beats the substance every time the substance is too expensive to check. Read it as the field manual for the Proxy Collapse.
The source
What do the Commonwealth Writers Prize AI allegations mean for prizes, and short stories? (The Conversation). Start with the scandal that triggered this edition, then notice how fast the debate became “is AI creative” instead of “what were we ever grading, and can we still grade it?”
Thanks for reading
Vlad