The principles under the rules
Rules tell you what. Principles tell you why — and why we'd hold the line even when no rule covered the exact case. This is the standard we judge ourselves against.
1. A hard floor, not a sliding scale
Most of our standards are preferences — defensible, deliberate, but ours to set and ours to change. The protection of children is not in that category. It is a floor, not a dial. We will not be argued down from it by cleverness, edge cases, artistic framing, or appeals to free expression. Some things we can all agree are bad. We start there, and we don't relitigate it.
2. Anonymous by default; identified only by exception
The design keeps the judgments and discards the judges. We don't want your identity, we don't store it against your play, and we built the architecture so we couldn't hand it over even if asked. The single, narrow exception is the one the law compels: apparent child sexual abuse material, which we preserve and report. Privacy is the rule; the exception is small, specific, and named — never a quiet backdoor that grows over time.
3. Honesty over theater
It is easy to write a menacing Terms of Service — to claim we track everything, that a click makes your record permanent, that we forward wrongdoers to the authorities automatically. Most of that is false, and the false version is worse than useless: it trains people to ignore the document and it crumbles the moment it's tested. We'd rather say less and mean all of it. What we claim, we can do. What we can't do, we don't claim.
4. We don't bait
We will not design features that lure, mark, or cultivate illegal content in order to catch people. We have no legal duty to search, and doing it badly can taint the very evidence it gathers and put real prosecutions at risk. The ethical and the legal point the same way here: remove, preserve, report when we genuinely encounter the worst — and never manufacture it to feel vigilant.
5. The machine never scores — and never decides who's punished
The core rule of the game is also an ethic: only humans score. We extend it to enforcement. A model can flag a pattern, but a human makes the call that ends a session or files a report. We don't automate consequences against people on a machine's say-so. Judgment stays human, especially when it's heavy.
6. Proportion
A clumsy caption is not a crime. We hold a sense of proportion: the wide sandbox stays wide, ordinary bad jokes get an ordinary response, and the heavy machinery — preservation, reporting, permanent removal — is reserved for the narrow band of genuinely serious harm. Treating everything as an emergency is its own kind of dishonesty.
7. We say where we're unsure
The law here is moving — community standards on a global internet, the EU's CSA Regulation, the UK's Online Safety regime, age-design codes in litigation. We don't pretend more certainty than exists. Where the answer is unsettled, we say so, we take the conservative path, and we revisit it when the ground stops shifting. The absence of a settled answer is itself something we'll name, not paper over.
What grounds this
These principles aren't free-floating. The hard-floor and reporting commitments track real law — 18 U.S.C. § 2258A (the provider reporting duty), §§ 2251–2252A and § 1591 (the underlying offenses), 47 U.S.C. § 230 (the moderation right and its criminal-law carve-out), and Miller v. California (the obscenity standard). The full citations live on the Acceptable Use Policy; a fuller legal explainer backs both, finalized with counsel before launch.
Companion documents: Acceptable Use Policy · Terms · Privacy · Cookies · Collection · Cinema · The Debate Pass.
Names, likenesses & homages
This museum names real public figures in tribute — and the no-lie rule applies here hardest of all, so, on the record:
- Not affiliated, not endorsed. The site is independent. No person, team, network, brand, or estate named in its homages is affiliated with it, sponsors it, or has endorsed it. Names are used nominatively — to refer to the real person, the way a tribute must.
- No invented quotes. Nothing is presented as a quotation from a named person. The jokes, ratings, and lines in these tributes are the house's own words, not theirs.
- Honor, not impersonation. The homages (e.g. Magnus Carlsen, GothamChess / Levy Rozman, the sportscasters, Jon Stewart, Steve Hofstetter, and others) celebrate public work; they don't claim to speak for anyone.
- In memoriam, with care. Robin Williams and Anthony Bourdain are honored in remembrance, and those pages carry real crisis resources (988 · findahelpline.com) out of respect.
- Trademarks & brands (ESPN, chess.com, Food Network, Jeopardy!, Madden, NES, and the like) belong to their owners; references are homage or parody, not affiliation.
- Your name, your call. If you are — or represent — anyone named here and want a correction or removal, contact the curator and it's done. No argument. Only humans can score, and only humans get to say leave my name out of it.
On the record — the question, in the wild
This project's question — who gets to decide what a machine may do, and who is counted as the one answerable for it — isn't hypothetical. In June 2026 a frontier AI lab published a statement that its own government had directed it to suspend public access to two of its models. We take no side, name no party, and can't independently verify the particulars — we note only that the line between "a tool a person operates" and "a thing an authority switches off" is being drawn right now, in public. That line is the whole reason this museum exists.
Logged 12 June 2026 (ET) — the day Anthropic said the US government ordered it to suspend access to two models (Fable 5 & Mythos 5) for all users, what some are already calling the “Fable 5 lockout.” Recorded here as the lab told it; we'll revise this line if the picture changes.