The Mesopotamian legal code governing water rights was inscribed during a period of relative rainfall. Two centuries later, officials were still pointing to it. The river was contracting. The code did not know.
The code had an author. You could, in principle, ask what the author intended. You could argue about whether the author's intent applied to new conditions. The author had left the room, but there was a room the author had been in — a context, a set of concerns, a reasoning that could be partially reconstructed from the text and its historical moment.
When an AI produces a document, the author-function has been separated from the text in a different and more troubling way.
The text exists. It may be well-structured, fluent, and substantively reasonable. But the claims it contains have no one who is responsible for their accuracy. No one who verified them against lived knowledge of the domain. No one who would be accountable if the text were acted upon and the action proved wrong.
This is not a problem with AI writing as a tool. The tool is what it is. The problem is with what happens after the tool is used: the document enters the archive, receives the authority of inscription, and begins to speak with the same formal confidence as any other document — without the accountability structure that should accompany it.
Institutions have spent centuries developing the mechanisms by which written authority is kept honest. Attribution. Peer review. Editorial oversight. Signatory responsibility. Version control. The ability to trace a document back to the person or persons who are answerable for what it says.
These mechanisms exist because authority without accountability is an unstable structure. It holds until someone acts on a document that turns out to be wrong, at which point the question becomes: who is responsible for the error? If the answer is genuinely no one — if the document was generated, accepted, distributed, and acted upon without anyone owning the claims it made — then the institution has a problem that is not primarily technical.
The question is not whether to use AI writing tools. The question is: when a document is produced with AI assistance and enters the institutional record, who is its author for governance purposes? Who has read and verified the claims? Whose name, whose standing, whose accountability is attached to the text?
If the answer is unclear, the document is not institutional knowledge. It is institutional noise wearing institutional clothing.
The archive will treat it with the same fidelity it treats everything else.
Someone has to decide whether that is acceptable.