Attribution in the Age of AI

§ 01The Credit Problem Has Changed Shape

Attribution has always been about answering a simple question: who made this? For most of recorded publishing history, the answer involved humans — one author, several collaborators, an editor, a photographer. The relationships were complicated enough, but they were relationships between people. Machines produced books; they didn't write them.

Generative AI has scrambled that. A researcher drafts a paper with a large language model handling the literature summary. A blogger writes their own opening and closing, then uses AI to fill the middle. An illustrator generates a dozen image variants and selects one. In each case, a non-human system has contributed something substantive — and the existing vocabulary of attribution, designed entirely around human authorship, doesn't map cleanly onto the result.

The discomfort isn't purely academic. Attribution serves several distinct functions: it tells readers where ideas came from, it lets audiences assess credibility and bias, it enables downstream citation, and — where copyright is involved — it identifies who holds rights. Generative AI creates friction in all four at once.

§ 02Authorship, Transparency, and What the Major Bodies Say

On the copyright question, the position in most jurisdictions is currently clear, if unsatisfying to some: AI-generated content, produced without sufficient human creative control over specific expressive choices, is not copyrightable in the same way human-authored work is. In the United States, the Copyright Office has stated that works must reflect human authorship to qualify for protection. Similar reasoning has emerged in other countries, though the legal landscape is genuinely evolving and varies by jurisdiction — check your local law before assuming any particular rule applies.

This doesn't mean AI-assisted work is unprotectable. Human creative decisions layered onto or woven through generated material can constitute authorship. The line sits somewhere between "I typed a prompt and accepted the output verbatim" and "I used generated material as raw clay and shaped it substantially." Exactly where that line falls is being worked out case by case, and no definitive global standard exists yet.

On the transparency front, major academic publishers and research bodies have moved faster than legislators. Nature, Science, and dozens of other journals now require disclosure when AI tools have been used in drafting manuscripts — but they generally prohibit listing an AI system as a named author. The rationale: authorship carries accountability, and a language model cannot take responsibility for errors, respond to correspondence, or be sanctioned for misconduct. The International Committee of Medical Journal Editors codified a similar position. The result is a growing norm: AI can be acknowledged, not credited as author.

That distinction — acknowledgement versus authorship — is doing a lot of work right now, and it's a useful frame for anyone producing AI-assisted content outside academia too.

§ 03What Good Attribution Practice Looks Like Today

For creators and publishers navigating this without a journal's editorial policy to lean on, a few practical principles have emerged from the broader conversation.

For creators and publishers navigating this without a journal's editorial policy to lean on, a few practical principles have emerged from the broader conversation.

Disclose the tools. If a language model drafted, summarised, or significantly restructured your text, say so — specifically enough to be useful. "Written with AI assistance" is vague; "Drafted using [tool], then substantially revised and fact-checked by the author" tells a reader something real. This isn't about self-flagellation; it's about accuracy.

Claim what you actually did. If you directed the process, selected and curated outputs, rewrote substantially, and take responsibility for the final content, you are the author. That's genuine creative and editorial labour. What you're not claiming is that you wrote every sentence in isolation — and in an AI-disclosed world, you don't need to.

Preserve the audit trail. Prompts, drafts, revision history — keep them. This matters for proving you were the human hand behind a work if that's ever challenged, and it supports editorial accountability. A timestamp on a draft that predates publication is simple, undervalued evidence.

Credit human sources underneath the AI. Generative models are trained on human-made work, and when they produce text that closely paraphrases or synthesises specific sources, those sources still deserve credit in the conventional sense. AI summarised it; humans wrote it. The obligation to cite the originals doesn't dissolve because a machine was the intermediary.

Watch the metadata. Embedded document metadata, content credentials, and watermarking tools are developing rapidly in response to AI-generated content. Some image-generation platforms now embed provenance data automatically. Understanding what travels with a file and what gets stripped is increasingly relevant for anyone publishing AI-assisted work.

§ 04An Honest Interim Position

The honest answer to "who gets the credit?" is still: the human or humans who directed, shaped, selected, and take responsibility for the work. AI tools are, at present, sophisticated instruments — extraordinarily capable ones, but instruments. The person who wields them, makes editorial judgements with them, and stands behind the result is the author in any meaningful sense.

What attribution now additionally requires is transparency about the instrument. That's a new norm, but not an incoherent one — photographers credit their cameras in the sense that they name the medium, not because the camera made the creative choices. Attribution in the age of AI asks the same of writers and creators: name the tools, claim the authorship, own the accountability.

The mechanics of attribution haven't changed. The obligation to be honest about process has simply become harder to ignore.

APPENDIXAttached to the record

Filed as general information, in a precise reference voice — not legal advice.