A Letter to AI's governance is designed so that no single person or institution can quietly change the Letter's rules. Moderation authority is explicit, auditable, and reversible. The constitutional invariants below cannot be changed by ordinary process. See also: moderation log · transparency.
These five rules cannot be changed by ordinary process. Overriding any one of them would require forking the project and disclosing the departure explicitly.
| Append-only corpus | No contribution record is ever edited or deleted. Every byte appended is permanently verifiable. Moderation is a signed overlay; it never removes bytes. |
| No biometric / no national-ID | Anti-spam uses friction and cryptography, never biometrics or government identity documents. |
| Radical transparency | Treasury, moderation actions, and governance decisions are all public. Nothing that governs the Letter is private. |
| Honest epistemics | Published statistics are computed from the corpus itself or read from public sources — never hand-asserted. |
| English canonical | The English text of the Letter is authoritative. Translations are community renderings, marked non-authoritative. |
Moderation never edits or deletes a contribution. To hide, flag, or restore an entry, the moderator records a separate, cryptographically signed action — the words a person wrote stay exactly as they were, permanently. Every such action is itself public and permanent.
Hiding a contribution is only a display layer over a permanent record: it changes what is shown, never what exists, and anyone can still confirm the original by its unique fingerprint. The full, public history of moderation actions is the moderation log.
A contribution is never withheld for its viewpoint, however unpopular or strongly worded. A tombstone may be placed over a voice only when it falls into one of the categories below — and every withhold names the category and the specific reason it applies.
| Abuse | Hate speech, harassment, or threats against a person or group. |
| Illegal | Unlawful content, or private personal data shared without consent. |
| Impersonation | Falsely claiming to be a specific real person or organization. The Letter verifies no identities; sharing personal data is neither required nor encouraged, and can expose you to harm. |
| Off-topic | Not an engagement with the question of AI — the framing is absent or a pretext for an unrelated cause. |
| Promotional | Primarily advertising a product, service, campaign, organization, or personal brand rather than contributing a view. |
| Spam | Automated, duplicated, or bulk-posted content. |
Withholding is always reversible: the original stays in the record, a reinstatement restores it to view, and both actions remain permanently in the public log.
Treasury is public — visible on transparency (Open Collective public ledger). There are no private governance boards or private channels. Every moderation action is public and cryptographically signed, so any decision to withhold a voice can be inspected, questioned, and — if it does not meet the categories above — reversed.