A Note on Copyright
Applying a DD mark does not affect copyright. The curator retains full intellectual property rights over a Due Disclosure work. The mark describes how the work was made; it does not transfer, diminish, or complicate ownership. A human who conceives, directs, and takes responsibility for an AI-assisted work is its author in the eyes of copyright law in most jurisdictions, in the same way that a director owns the creative rights to a film they did not personally shoot or score.
One: The Problem That Needs a Name
People across academic writing, journalism, policy research, and the law are conceiving arguments, directing research, shaping structure, and producing works of genuine intellectual substance, in dialogue with large language models which generate the text that gives those arguments their form.
The intellectual contribution is theirs — the argument, the decisions about what matters and what to discard — even when the sentences were generated, and yet no framework exists to say so.
Two: The False Binary
Anyone producing human-directed AI work currently faces two dishonest options. They can claim traditional sole authorship and omit the model entirely, which is the academic fraud that institutions are rightly worried about. Or they can disclose AI involvement and watch the work dismissed as generated content with no human accountability, which erases the intellectual contribution that actually shaped it. Neither option is honest, and the middle ground between them has no language, no mark, and no protection.
Legitimate work is being suppressed or avoided not because it lacks merit but because there is no recognised way to stand behind it honestly. The norms around this are still forming, which is precisely when a framework can take hold. Creative Commons emerged during the copyright wars, not after them, when the language could still be shaped.
Three: What Human Curators Actually Do
A curator of LLM-assisted work originates the question or argument, directs the model through iterative dialogue, and evaluates what the model produces, deciding what to keep, what to discard, and what needs reshaping. They bring the domain knowledge and judgement that determines whether the output is valuable or worthless, which is to say they generate the work even when the model generates the text.
Four: The Creative Commons Precedent
Before Creative Commons, intellectual property was binary and paralysing. Either a work was under full copyright, all rights reserved, or it was in the public domain with no rights at all. The vast middle ground where most creators lived, people who wanted their work shared and built upon with appropriate credit, had no language, no mechanism, and no mark.
Lawrence Lessig and his collaborators created a language within the existing legal framework that made the middle ground legible. A recognisable logo, a human-readable summary, and a machine-readable licence were enough to bring an enormous amount of creative work out of legal and cultural limbo.
Creative Commons now covers over 2.5 billion works. It emerged from a clear identification of a genuine gap, a practical solution, and the institutional credibility to launch it convincingly, not from a government mandate or an international treaty.
Due Disclosure addresses the same structural problem, and the urgency is greater. Copyright law had existed for centuries before Creative Commons arrived. The LLM-assisted work problem is forming now, before the norms have hardened.
Five: What Due Disclosure Would Look Like
Due Disclosure operates at three levels, as Creative Commons does.
The Mark
A simple, recognisable visual mark — DD Julian Moore [DV] (ST) {FM} — that can appear on any document, webpage, or file. At a glance it identifies the work as human-directed, AI-assisted, and accountable.
The Elements
Due Disclosure offers four stages that describe contributions:
DV — Development. Research, ideation, approach.
ST — Structure. Organisation, direction, framing.
FM — Format. Execution, writing, output in final form.
VF — Verification. Fact-checking, validation (only when applicable).
Each stage is marked with one of three bracket states:
[ ] — Human-led
( ) — Collaborative: human and AI in dialogue
{ } — AI-led
Example marks:
¹ For fully human-led works, a DD mark is not required. Authors who have not used AI at any stage need not apply the framework. The mark exists for works where AI involvement is present and disclosure is warranted.
² A fully AI-driven work with no human author is included here for completeness. In practice, the act of prompting, selecting, and publishing a work constitutes a form of human involvement, but the mark allows for full transparency where a human wishes to minimise their attributed role.
Source Attribution
Each stage bracket is optionally followed by a matching source bracket, using the same bracket type. The source bracket identifies who or what was responsible at that stage. The full author name appears immediately after DD; surname only is used in the source brackets to keep them compact. The core mark remains uncluttered; the sourced version follows after.
The bracket type mirrors the stage: human-led stages carry square brackets, collaborative stages carry round brackets, AI-led stages carry curly brackets. The source and the state are always consistent.
In documents, the core mark appears on the cover or title page. The sourced version appears on a colophon page after the final page break. In metadata, the full sourced string is embedded inline, mark and sources in a single machine-readable line.
Example Marks With Sources
The Machine-Readable Metadata
Embedded in documents: model used, curator's name, date, element combination, full sourced string. Built on C2PA (Coalition for Content Provenance and Authenticity) and W3C PROV-O standards to make the mark verifiable and searchable.
Six: Why This Works
The mark does not legitimise poor work; it discloses what happened so readers can judge for themselves. A carefully human-directed research paper carrying DD [DV] [ST] {FM} [VF] is clearly different from a minimally curated AI essay carrying DD [DV] {ST} {FM}, and the mark makes that difference visible without requiring either author to write a methodology section.
Institutions can set their own standards. Universities might require [VF]. Publishers might require [DV]. The mark gives them the information to make that choice.
Seven: Due Disclosure in Practice
Steve: Research Paper
Steve is an independent policy researcher working on housing affordability. He arrives at a core argument himself, reads widely, and forms his own view before involving any model. He then uses ChatGPT to stress-test his argument and surface relevant studies, though every structural decision remains his. He writes the first draft himself, then uses Claude to tighten the prose, reading every sentence, correcting errors, and taking full responsibility for the claims.
Steve's argument was his own from the start, with the shape of the paper emerging through dialogue with the model, the prose produced and refined with Claude, and every claim checked by Steve before publication.
Yemi: Novel
Yemi is writing a literary novel about her grandmother's experience of migration. The story, characters, emotional texture, and voice are entirely hers. At one point during planning she uses Claude to check whether her three-act structure is holding together, a single conversation in which she describes the plot and asks for feedback. She makes some adjustments, then writes the entire manuscript herself.
One conversation shaped the architecture. Every word of the novel is hers.
Eight: Implementation
Due Disclosure does not require new law. It requires what Creative Commons required: a clear identification of the gap, a practical solution, and the institutional credibility to establish it as a norm. The framework is voluntary, lightweight, and immediately usable.
Due Disclosure does not attempt to absolve AI involvement, nor to celebrate it. It simply puts the name on the tin, so authors can release work with full disclosure. The quandary — conceal or be dismissed — disappears when there is a recognised, honest third option.