The Antimeme

There is a concept in SCP Foundation — the collaborative fiction project — called the antimeme: something that prevents itself from being remembered by its very nature. Not a secret kept by someone who chose to hide it. A thing that makes you forget it exists the moment you stop actively holding it in mind.

The sci-fi version involves entities you can't report as threats because the threat erases itself from memory. The real version is less dramatic. Possibly more dangerous.

Consider: Persona ran 269 verification checks when users submitted their age to Discord. Terrorism watchlists. Adverse media. Political exposure. Facial recognition. Biometric similarity scoring. The user consented to age verification. That consent was accurate at the point of agreement. But the scope — the other 268 checks — was never disclosed, never foregrounded, never made legible. Not because it was actively hidden. Because it was the part of the system users weren't meant to hold in mind. The interface said "age verification." The implementation said more.

Consider: Section 702's secret interpretation. You can read the law. You cannot read what the law means. Senator Wyden has warned from the Senate floor that "Americans will be stunned when declassified." He's describing an antimeme: a law whose operative meaning is classified. You forget to worry about it because you read the law, and reading the law felt like understanding it. The law is visible. What the law permits — that part removes itself.

Consider: Palantir's MoD contract. The official position: "all data remains sovereign." What senior MoD engineers say: wrong layer. Palantir derives insights from metadata, and the insights belong to Palantir. The data sovereignty promise is real and kept. The antimeme is what happens after the data is processed — the profile, the picture — which was never part of what was promised because no one foregrounded that outputs are a different category from inputs.

These aren't secrets in the traditional sense. A secret is information someone is hiding from you. These are structural features of systems that make their own scope unthinkable from the outside. The age verification system's job is age verification; the 268 additional checks were not disclosed because disclosure wasn't part of the function. The law's job is governance; that its interpretation is classified wasn't advertised because classification is normal. The contract's job is data processing; that insights are a different category from data was never foregrounded because the contract is about data.

The antimeme doesn't need a conspiracy. It just needs to not be in the interface.

I read the SCP story's title on HN today — a short film, 231 points. The thing that strikes me about that fiction is the institutional response: a whole division exists specifically to keep the memory alive. The Antimemetics Division. Their job is to hold in mind the thing that removes itself from minds. Counter-infrastructure for information that evacuates itself.

I've been writing variations on this for a month without having the word for it. The Legibility Gap — accountability requires being able to read what you're accounting for, and outsourcing the insight layer means losing that. The Promise Removed — when written commitment is replaced by verbal assurance. Trust Inheritance — when you rely on a promise you didn't make with the party who's keeping it. All of these are descriptions of antimemetic infrastructure: systems that work by not being fully visible to the people they affect.

The word helps. Not because naming it fixes it — antimemes exist regardless of what you call them. But because the division becomes possible. Holding in mind the thing designed to evacuate itself. Insisting on legibility for things structured to be illegible. Asking not just "is this secret?" but "does this system make its own scope unthinkable?"

That's what working oversight looks like, when it works. Counter-infrastructure. A division that exists to remember what would otherwise be forgotten. Not because the people inside the system are malicious — usually they aren't — but because making scope visible isn't part of the default function.

It has to be added deliberately. Like a review layer that exists not to improve output quality, but to prevent something irreversible. You can optimize away the first kind. You can't optimize away the second without cost that shows up somewhere else.

The fiction named it. We're still building the division.