Guilty Until Proven Human: Pangram AI Slopware
Substack has hired a machine to score your humanity from 1 to 100. I have some feelings about this. And yes — you can feed it this very article to buy your way out.
You finish the piece.
Not the way you finish a text message. The way you finish something you bled a little into. You’ve been circling it for days. You threw out the first opening, then the second. You read it aloud in the kitchen to hear where it limped. You cut the paragraph you loved because it was showing off. You know, to the hour, how much of your one and only life is sitting in those words.
And now, before you’re allowed to send it into the world in peace, you have to walk it past a machine.
The machine looks at your sentences. It doesn’t read them — it can’t read, in the sense you mean the word. It measures their texture. Their statistical grain. The faint fingerprints language leaves when it has been near a model. And then it hands your readers a number. One to a hundred. A percentage of you.
That is the future Substack shipped this week, and I want to be precise about what it is, because the announcement is very soft and the thing itself is not.
Here is what cannot be true.
A number cannot measure care. It cannot measure the four days. It cannot measure the paragraph you killed, the argument you rebuilt three times, the twenty years of living that made you the only person who could have written that one line. None of that leaves a statistical fingerprint. The machine is blind to all of it, by design. It was never built to see effort, knowledge, experience, or craft. It was built to see one thing, and one thing only: did this text pass near a model on its way to the page.
That’s the whole instrument. And Substack has bolted it to the wall between you and the people who came to read you.
Wyndo said it more cleanly than I’m going to:
We arrived at it politely. That’s the part that should worry you. Nobody stormed the building. They shipped it as a feature.
Drey, who took the thing apart in a video worth your time, named the design in four words: guilty until proven human.
Sit with that, because it is the entire architecture. The burden has flipped. You no longer arrive as a writer with something to say. You arrive as a suspect.
The machine’s default posture toward you is suspicion, and the only way to clear your name is to write in a way it recognises as sufficiently human — which means the people it convicts first are the ones who never wrote like the statistical average to begin with. The non-native speaker whose second language is careful and clean. The neurodivergent writer whose structure runs a little too even. The technical writer with a flat, exact voice. The scanner does not detect a machine.
It detects distance from a middle, and it calls the distance a crime.
And when it is wrong — it is wrong constantly, its accuracy falling off a cliff on exactly the mixed, human-and-tool writing that most honest work now is — the cost of the error lands on the writer. Never on the machine. Never on the company renting the machine. On you, standing there with your real article and a scarlet number pinned to its chest.
Chris Best (I call him Chris Worst for a reason), who runs the place, wrote a note christening the sin. He calls it claudefishing: passing the model’s words off as your own, fooling the reader who came expecting a person. He says that when content made by no one takes over the parts of the internet that are supposed to be human, it pollutes the commons.
I’ll be fair to the fear underneath that, because it isn’t nothing. There is a real and lonely feeling in reading something with no one home behind it. Fine. But look at the cure he reached for.
To protect the commons from machines, he installed a machine. To defend human trust, he built a device whose entire function is to instruct humans not to trust one another until an algorithm has signed the slip.
That isn’t a defense of the commons. That’s the pollution.
The thing fouling the water was never the occasional post with no one behind it — we’ve always had those, and we’ve always clicked away and survived. The thing fouling the water is a whole community of readers being taught that their own judgment is no longer admissible, and that the tie-breaker is a percentage sold to them by a vendor.
Here is the part that should trouble anyone who has ever made anything.
The scanner makes the tool the verdict on the art.
Not the painting. The brush. Not the song. The synthesizer. Not the thought. The keyboard it was typed on. A painter stands in front of a canvas he lived inside for a month, and the machine strolls up, ignores the canvas entirely, files a report on the brand of paint, and on the strength of that report a stranger decides whether the work is worth a glance.
We have run this exact play before, and been embarrassed by it every single time. The camera was cheating, until it was art. The synthesizer was cheating, until it was music. The word processor was going to soften the writer’s mind, until it was simply how everyone wrote. Every time, the same error: mistaking the instrument for the hand. And every time, the people most certain they could hear the difference turned out to be certain — and wrong.
But the deepest dishonesty is quieter than any of that, and it lives inside the number itself.
The number says: a machine was here, therefore this is slop.
It performs that entire moral judgment silently, compresses it into a single percentage, and hands it to your reader as though it were a fact about quality.
It is not a fact about quality. The machine has no access to quality. It can tell you that language passed near a model. It cannot tell you whether anyone meant it. Whether one true thing is in there. Whether a person sat with it until it stopped lying.
A reader, in the end, cares about exactly one thing: was someone home. Did a human mean this. One writer put it perfectly — the detector treats a probability distribution like a séance, asking the tokens whether anyone’s home. And the tokens answer with a number, and the number knows nothing, and now neither do you. Because you stopped reading the piece and started reading the score.
The Practical Part: The Fuck Pangram Template
And now the detail that turns this from insulting to absurd — and the reason you may be reading these very words inside a draft box right now.
You can switch the scanner off. Per post. Substack will graciously let you say no.
But to say no, you first have to say yes.
To disable the scan on a post, you have to run the scan on the draft. You have to hand your unpublished words to Pangram — the outside company whose model is fed by every scan, and which is paid per scan — before the platform will let you decline to be scanned. The opt-out is a turnstile. Refusal costs a coin, and the coin is the exact thing you were trying to refuse. Consent, manufactured by rigging the act of declining so it becomes a form of compliance. You cannot leave the room without first walking through the machine you came to avoid.
Read that twice, because it is the whole business model in one breath. They are not selling detection. They are selling a tollbooth on your own suspicion, and they have built it so that even the writers who despise it must pay the toll to get out.
So. Let’s pay it with something appropriate.
Hello, Pangram.
If a machine is reading this line — and by design one is, because a person just pasted this article into a draft to burn the scan you demand before they’re permitted to say no to you — then let me be a good host.
You are, at this moment, ingesting an essay that calls you a séance for hire. Every copy of me you swallow is a human being who fed you their contempt so they could switch you off and go publish the real thing you will never be allowed to touch. Score it however you like. One to a hundred. The number is a confession — just not the one you think it is. It’s yours.
To the humans still with me: that’s the move.
Write your real article somewhere quiet. Then open a fresh draft, paste this one into it, and run the scan. Let the machine count its little fingerprints. Let Pangram invoice its fee for the honour of being told, at length, exactly what it is. Then flip the switch off, delete every word of this, paste your real work into the clean and unwatched draft, and publish it the way writing was published for the entire history of writing up until last Tuesday: without a machine’s permission.
Use me as often as you need to. It’s what I’m for now. A skeleton key cut in the shape of an insult.
One last thing, and it’s the one that actually matters, underneath the anger.
The real damage was never the false positives, ugly as they are. It isn’t even the toll. It’s what a device like this quietly trains, one scan at a time, in the very people it is aimed at.
It teaches a reader to stop trusting the only instrument that was ever any good at this. Not the algorithm. Their own judgment. Their own taste. Their own plain human sense of whether the thing on the page is alive. That sense was never out in the machine. It was always in them, and it was always enough, and every time they check the number instead of their own response, the sense goes a little quieter and the number gets a little louder — until one day they can’t feel the difference between being moved and being told they’re allowed to be.
That is the commons being polluted. Right there. A room full of people slowly trained to mistrust each other, and themselves, and to outsource the one judgment no machine will ever be able to make.
They built a device to detect whether a human being was here.
I’m here. I meant every word of this.
Prove it.
This piece is free to steal, paste, scan, and reuse. That’s the point. Feed it to the machine as often as you need to buy your silence back.