The Problem Is Not What You Think
The common framing of AI and authorship centers on quality, credit, and labor. Can an AI-generated novel be nominated for a literary prize? Should authors disclose AI assistance? Who owns the copyright to machine-generated text? These are real tensions, but they are not the provenance problem. The provenance problem is simpler and harder: how does anyone know when a piece of writing was created?
Before language models, the question of document timing was relatively straightforward. You could examine file metadata. You could look at revision histories. You could consult the author's email records. These were imperfect — easily manipulated by a sophisticated actor — but they were good enough for most purposes, because tampering required both technical capability and motive, and the motive was usually obvious after the fact.
Language models change the cost structure. A person can now produce a plausible novel-length manuscript in minutes, upload it to a file system with a custom creation date, and present it as having been written over three years. The output may be excellent. The fabricated timeline costs nothing. The motive is frequently obvious — priority claims, copyright disputes, fraud — but the verification, without independent evidence, is impossible.
The Authorship Crisis That Already Happened
The authorship crisis is not coming. It is already embedded in the content that has been published since 2023. An unknown fraction of books, articles, academic papers, and blog posts published in the last three years were generated, in whole or in part, by language models. The fraction is unknown because detection is unreliable.
AI detection tools — the various classifiers trained to identify machine-generated text — have false positive rates that make them legally and editorially unusable. A human author with clean, direct prose may score as AI-generated. A verbose language model output may score as human. The tools are improving, but they are fundamentally solving the wrong problem: they are trying to determine authorship from text properties rather than from external, verifiable provenance records.
The question is not "does this text read like a human wrote it." The question is "does a dated, tamper-proof record exist showing that a specific author possessed this specific text at a specific moment in time." Those are different questions, and only the second one is answerable.
What Provenance Is — And Is Not
Provenance is a chain of custody. In art, it is the documented ownership history of a work, from creation to the present. A painting without provenance may be genuine, but its value and tradability are limited because its history cannot be verified. A painting with clear provenance — receipts, exhibition records, credible ownership documentation — can be assessed as authentic not because anyone can prove Vermeer held the brush, but because the documentary chain is unbroken and consistent.
Literary provenance is the same idea applied to text. A manuscript with provenance has a verifiable creation record that shows: who created it, when, and whether it has been modified since. Traditional literary provenance relies on editorial records, agent correspondence, publisher logs — all mutable institutional records.
Cryptographic provenance replaces mutable institutional records with immutable mathematical ones. The text is hashed. The hash is anchored on a public blockchain. The blockchain timestamps are not controlled by any single party and cannot be retroactively altered.
This does not prove the author is human. It does not prove the text is original. It proves that the specific text, in its current form, existed at the time of anchoring — and that the anchoring date cannot be fabricated after the fact.
Why This Matters Right Now
Priority disputes in literary publishing — who wrote what first — have always been difficult to resolve when they span years. The journal article you submitted in 2021 and the book you published in 2023 may share ideas with a manuscript that appears in 2025 claiming priority from 2019. Without comprehensive provenance records, these disputes resolve through institutional reputation and legal resources, not through evidence.
In the AI era, the priority dispute problem compounds: a sophisticated actor can produce a manuscript that anticipates your ideas — using your published work as training data — and timestamp it before your own publication. This is not a distant risk. The capacity to do it exists today. The incentive structures (plagiarism claims, derivative work disputes, fraud) exist today. The institutional mechanisms to detect it do not.
The practical solution is not to hope the problem doesn't arrive. It is to create a public, verifiable provenance record before it does. An anchoring timestamped today cannot be contested by an anchor that appears tomorrow claiming to be from yesterday.
The Irony of the Deal Chain
The 2,500 Donkeys was written about a world where documents are treated as proof without any verification mechanism. The PPP deal chain runs entirely on documents that everyone receives, no one verifies, and everyone forwards. The BCL, the MT799, the NCND — these documents look like proof. They establish the form and language of financial legitimacy. They provide no actual verification.
The deal chain and the AI authorship problem are structurally identical: they both exploit the gap between the appearance of a legitimate record and the actual verifiability of that record. In both cases, the document looks right, the timestamps look right, and the institutional formatting looks right — because those things are easy to fake.
Blockchain anchoring addresses the gap. Not by making fakery harder — it was already hard to fake SWIFT certificates, and people faked them anyway — but by creating a verification record that doesn't depend on trust. The hash is either correct or it isn't. The block timestamp is either real or it doesn't exist. The truth is in the math.
The book was written by a person, in Norcross, GA, across 2023 and 2024, and anchored on Polygon Mainnet before its public release. The chain of custody is on-chain. The deal went through. The 2,500 Donkeys is the receipt.