RatedWithAI

RatedWithAI

Accessibility scanner

AI CopyrightSeptember 22, 2026

Your AI Registration Was Refused. Now There Is a Clock.

Nearly everything written about AI and copyright is about whether a work can be owned. This is about the letter that says yours cannot — what the Office will reconsider, in what order, by when, and why the answer decides whether you can ever sue anyone over it.

The refusal is not the end of the claim; the calendar is. Two administrative reconsiderations stand between a refusal and a court, each on a three-month clock. Use them and your original filing date survives. Abandon and refile and you have traded the date away for a form that felt easier to fill in.

What the Clock Actually Says

The refusal letter starts a three-month window, not an open-ended one

An applicant who disagrees with a refusal may ask the Office to reconsider, and the request has to arrive within three months of the date on the refusal letter. That date is on the letter, not on the day somebody in your team opened the email. Miss it and the administrative route closes; the Office's regulations treat a late request as one it will not entertain.

There are two reconsiderations, and you must use both

The first request goes back to the registration program that issued the refusal — usually to a different, more senior examiner. If that is refused too, a second request goes to the Review Board, which is the Office's final administrative word. Each has its own three-month clock running from the previous refusal, and each carries its own fee.

The Board's decision is the end of the agency road

A Review Board decision is final agency action. That is the point at which a court can be asked to look at it, and it is also the point at which the record is fixed: what the Board saw is substantially what a reviewing court will be handed. Arguments invented afterwards arrive late.

Your filing date does not move

This is the fact that changes the arithmetic of whether to appeal. The effective date of a registration is the day the Office received an acceptable application, deposit and fee — not the day it finally decided. A claim that registers after two reconsiderations carries the original date. Abandon the application and refile later, and you have bought a new, worse date.

Read the Refusal for Which Problem You Have

Four grounds do most of the work, and they need four different answers. The appeals that go badly are the ones that answer the subject instead of the sentence.

The claim was broader than the human contribution

The most common shape of an AI refusal is not 'this work is unprotectable' but 'this claim covers more than you authored'. If the application claims the whole image, the whole article or the whole track, and the Office believes a model generated the expressive content, the claim as filed fails even where a narrower one would have registered.

The AI material was not disclosed

The Office asks applicants to identify AI-generated material that is more than de minimis and to exclude it from the claim, describing the human authorship that remains. An application that says nothing, and a work whose provenance later becomes obvious, is a much worse file than one that disclosed and disclaimed at the outset.

Prompts were offered as the authorship

The Office's published position is that prompting, by itself, does not generally make the prompter the author of what comes back, because the prompt does not control the expression in the output the way an author controls a brushstroke or a sentence. An appeal built entirely on the length, craft or iteration count of the prompts is arguing against the Office's stated view rather than around it.

The selection-and-arrangement claim was never actually made

Where a human chose, edited, ordered and modified model output, there is often a real claim in the selection, arrangement and the human modifications — but only if the application says so and the deposit shows it. Many refusals are refusals of a claim nobody bothered to draft.

Building the Record You Will Be Stuck With

Reconstruct provenance before you write anything

Which model, which version, what went in, what came back, and what a person did to it afterwards. Original files with timestamps, editing history, layered source documents, version control, drafts. This is evidence of human authorship and it is the only kind that carries weight; a declaration that a lot of work went in, unaccompanied by artefacts, is not persuasive to an examiner who has read a thousand of them.

Decide what you are actually claiming, then claim only that

Narrowing is not surrender. A registration limited to the human-authored text, the human photography, the human-drawn elements, or the selection and arrangement of the whole, is a registration — and it is the one that survives an infringement defendant's inevitable attack on validity. A refused broad claim protects nothing at all.

Answer the refusal's reasoning, not the topic

Read the refusal for the specific ground: no human authorship, de minimis authorship, undisclosed AI material, or a deposit that does not match the claim. Each has a different answer. A general essay on creativity and machines does not respond to a deposit mismatch, and the second reconsideration is not the place to discover which problem you had.

Fix the disclosure honestly, including retrospectively

If the original application failed to disclose AI material, correcting it in the reconsideration is far better than leaving it. Inaccurate information supplied knowingly, and material to the decision to register, is the hook a later infringement defendant uses to attack the registration itself — a problem that outlives the examination by years.

Keep the commercial clock in view

The reason to run this process rather than shrug at it is downstream: registration is the gate to filing an infringement suit over a US work, and the timing rules that unlock statutory damages and attorney's fees are keyed to when you registered relative to publication and to the infringement. A refusal you never appealed is a work you sue over, if at all, with actual damages only.

Five Ways This Goes Wrong

Refiling instead of appealing

It feels faster and it silently discards the effective date. Where the earlier date matters — and it usually does, because infringement rarely waits — reconsideration is the cheaper instrument even though it takes longer.

Treating the examiner's letter as a negotiation

The reconsideration is a written submission into a record. Phone calls and informal correspondence do not substitute for it, and the three-month clock does not pause while a conversation happens.

Claiming the model's output as work made for hire

Work made for hire allocates authorship between humans and the entities that employ or commission them. It does not manufacture human authorship where the Office has found none, and an appeal that leans on it is answering a question the refusal did not ask.

Ignoring the vendor's terms

A generative vendor's terms may assign you whatever rights it has in the output, and that assignment is worth exactly whatever those rights turn out to be. An assignment of nothing is still an assignment. Read the contract for the indemnity and the warranty, not for a copyright the vendor cannot confer.

Shipping the work as though the refusal changed nothing

A refused claim is a business fact, not just a paperwork fact: it affects what you can license exclusively, what you can promise a customer in an IP warranty, and what your own contracts say you own. Somebody outside the legal team needs to hear it.

Questions Applicants Ask

How long do we have to appeal a copyright refusal?

Three months from the date on the refusal letter for the first request for reconsideration, and three months from the date of that second refusal for the request to the Review Board. Both dates run from the Office's letter, not from internal circulation, and both requests carry their own fee. The practical failure is almost never a legal one — it is that a refusal letter lands in a shared mailbox in week one, reaches the person who can decide anything in week eleven, and reaches outside counsel after the window has closed. If your organisation files registrations at any volume, the cheap fix is a rule that any letter from the Office is routed the day it arrives and the deadline is diarised on receipt.

If we appeal and eventually win, is our registration dated from the original filing?

Yes, and this is the single strongest reason to use the process rather than abandon and refile. The effective date of a registration is the day the Copyright Office received an application, deposit and fee that it ultimately finds acceptable — not the day examination concluded. A claim that takes two reconsiderations and many months to resolve still bears the original date once it registers. Refiling a narrower claim later does the opposite: it gives you a fresh, later date, which can move you outside the windows that matter for statutory damages and attorney's fees, and which hands an infringement defendant an argument about everything that happened in between.

Should we narrow the claim rather than fight for the whole work?

Usually, yes — and it is worth being clear that narrowing is a different thing from losing. Where a person selected, arranged, edited and modified model output, there is frequently a genuine claim in those human contributions and in the selection and arrangement of the whole, even though the raw generated expression is disclaimed. A registration on that narrower basis is enforceable, survives a validity attack, and can be obtained now. A broad claim that the Office will not accept protects nothing while you argue about it. The exception is the case where the human contribution really is the expressive core and the model was a tool in the ordinary sense; that is worth arguing, and it is worth arguing with files and version history rather than adjectives.

We never disclosed that AI was involved. What now?

Correct it, and correct it in the reconsideration rather than hoping the question goes away. The Office expects applicants to identify AI-generated material that is more than trivial and to exclude it from the claim while describing the human authorship that remains. The reason to fix an omission now is not tidiness: a registration obtained on information that was inaccurate, supplied knowingly, and material to the decision to register is a registration an infringement defendant will attack directly, and that attack is far more damaging than a narrowed claim would have been. Supplementary filings and corrections exist for exactly this. Treat the disclosure as part of the asset, not as a disclosure risk to be managed.

Can we skip the reconsiderations and go straight to court?

Not if you want a court to review the refusal itself. Judicial review of an agency decision generally requires exhausting the agency's own review process, and here that means both requests for reconsideration and a Review Board decision. There is a separate and often-confused route: where registration has been refused, a copyright owner can still bring an infringement action provided the statutory notice is served on the Register, who may intervene on the registrability question. That route puts registrability in front of a judge as part of your infringement case rather than as a challenge to the refusal, and it is not a reason to let the administrative clock lapse — it is a parallel mechanism with its own procedural conditions and its own consequences for remedies.

The Provenance Test

Pick one asset your company published this quarter that a model touched. Ask whoever made it for the file history: which model, which version, what the human did afterwards, and in what order. Give them an afternoon.

If it comes back, you can draft a claim that will register. If it does not, you have learned that on your own schedule rather than three months into a reconsideration — and the cheapest day to start keeping that history is the day before you need it.

Related Reading