What Matters
Paying for an AI tool does not settle who owns copyright in its output. For a US-facing release, separate permission to use the service, protectable human expression, the chain of ownership and other parties’ rights. DN’s evidence record helps teams document those questions before promising exclusivity to a customer. Its counts describe documentation gaps, never a legal approval.
DN Evidence Block
Prepared for Decentralised News · Editorial by Heath Muchena · Primary-source review and original DN instrument; independent field testing is disclosed only where performed.
| Evidence reviewed | 11 November 2026. US federal copyright law, official Copyright Office materials, the Thaler appellate opinion and Supreme Court docket. |
|---|---|
| Jurisdiction and decision | US copyright evidence for a particular AI-assisted asset before commercial release or a customer ownership promise. |
| Original DN instrument | DN AI Copyright Evidence Record v1.0: seven evidence areas, unresolved questions and a reusable JSON export. |
| Evidence type | Primary-source legal research and an original documentation workflow. Examples are invented teaching scenarios. |
| Limits | No court outcome prediction, rights clearance, provider ranking or legal review of readers’ files. Patents, trademarks, publicity, privacy and non-US rules require separate analysis. |
The Signal: An Output File Is Only the Beginning of the Rights Record
Imagine an agency delivers a landing page with generated illustrations, edited copy and a small software component. The customer asks for exclusive ownership. The agency has a subscription receipt and a final ZIP file. Those documents show a purchase and a delivery, but leave the central questions open: which expression came from a person, who contributed it, what was licensed into the project and what rights are actually being promised?
The practical unit of review is the asset and its components. A page, illustration, soundtrack and source-code file can have different histories. A single statement that the project was “made with AI” is too coarse to support a useful commercial decision.
This article examines that evidence problem. DN’s existing agent rights and responsibilities analysis addresses authority and accountability more broadly. The narrower question here is what a team should be able to show when its customer asks who owns an AI-assisted deliverable.
Claim Versus the Evidence You Can Actually Check
The claims below are common decision shortcuts used as teaching examples, rather than quotations attributed to a particular vendor. The evidence column identifies the relevant record or legal source.
| Claim being tested | Relevant evidence | Decision consequence |
|---|---|---|
| “We paid for the output, so we own its copyright.” | Copyright initially belongs to the author, subject to rules including work made for hire; ownership can be transferred. 17 USC §201. | Keep the subscription purchase and the ownership chain as separate records. |
| “Our prompts make us the author of every generated element.” | The Copyright Office’s Part 2 report evaluates human control over expression and says prompts alone are insufficient under the technology it examined. USCO Part 2. | Identify the expression contributed by a person, with supporting versions. |
| “The contractor was paid, so all rights transferred.” | Transfers of copyright ownership generally require a signed writing unless they occur by operation of law. 17 USC §204. | Check the actual agreement and the rights it covers. |
| “A copyright notice resolves the asset’s provenance.” | A notice identifies a claim; the evidence still needs to establish its basis. USCO Copyright Basics. | Retain provenance independently of the label printed on the work. |
| “Seven completed fields mean the asset is legally cleared.” | The DN instrument checks entered fields and references only. | Treat completeness as an administrative fact. A qualified reviewer still evaluates the evidence and proposed use. |
What the Current US Sources Establish
In Thaler v. Perlmutter, the D.C. Circuit affirmed the refusal to register an artwork whose application identified a machine as its sole author. The 18 March 2025 opinion holds that eligible work must originate with a human author. It did not decide the full boundary of human contribution in every AI-assisted workflow; Thaler’s alternative claim to authorship through creating and using the machine was waived before the agency. Read the appellate opinion.
The Supreme Court’s official docket records that the petition for review was denied on 2 March 2026. That is the recorded procedural outcome; it should not be described as a new Supreme Court merits ruling about every kind of AI output. Docket 25-449.
The Copyright Office’s 2025 report distinguishes purely generated material from protectable human contributions, including qualifying expressive inputs, creative modifications and creative selection or arrangement. It treats sufficient authorship as a fact-specific question. That supports a component-level review, rather than a blanket conclusion that every AI-assisted project has identical protection. Part 2, executive summary and authorship analysis.
DN’s practical inference is to preserve the creation history while the work is being made. Reconstructing it after a dispute will often be harder than retaining meaningful versions and a short contribution note at the time of production.
Four Questions to Keep Separate
1. What does the service agreement permit?
Record the provider, product, plan and terms version that applied to the work. Identify any permitted-use language and relevant restrictions, then keep the reference with the asset. Do not assume that another product from the same company uses the same agreement. This article does not compare current vendor contracts.
A contractual permission answers a question about the parties’ agreement. DN recommends recording it separately from the statutory question of whether a particular expression attracts copyright. A promise can only be assessed against its actual wording and scope.
2. Which expression is the copyright claim about?
Identify specific text, image regions, arrangement or code changes. Section 102 protects qualifying original expression, while excluding ideas, processes, systems and methods of operation from copyright protection. A product concept and the expression implementing it therefore need different descriptions in the record. 17 USC §102.
3. How did the relevant rights reach the business?
Start with the contributors and the applicable ownership route. The Copyright Office explains that work made for hire has defined conditions: employee work within employment scope, or specified categories of commissioned work with the required agreement. Paying a freelancer is not a complete substitute for that analysis. USCO Circular 30.
Where the business relies on an assignment, retain the signed instrument and its asset scope. Section 204’s writing requirement concerns transfers of ownership; avoid applying that statement indiscriminately to every nonexclusive permission. 17 USC §204; USCO Copyright Basics.
4. What other rights or restrictions affect release?
Record supplied photographs, audio, fonts, code, datasets and other components. A conclusion about your contribution does not answer whether another party’s material was used permissibly. The Copyright Office describes permission and statutory limitations as distinct routes relevant to using existing copyrighted work. USCO Copyright Basics.
For this workflow, similarity review is a recorded task with a stated scope. Its existence is not proof that infringement is impossible. The person responsible for release should know what was examined, what remains unresolved and which proposed customer promises depend on those answers.
The DN AI Copyright Evidence Record
The record uses seven evidence areas. Each can be marked reference recorded, unresolved issue or not recorded. A reference is a non-sensitive identifier for material your team retains elsewhere. The tool does not fetch documents, authenticate an agreement or decide whether a claimed human contribution qualifies for protection.
| Evidence area | What the record should point to |
|---|---|
| Asset and component identity | A fixed deliverable version and an inventory separating generated, human-created and externally supplied components. |
| Provider terms | The applicable product, plan, terms date and use/rights provisions. |
| Input permissions | The source and relevant permissions or other assessed basis for supplied material. |
| Human contribution | A specific contribution description and meaningful before/after versions. |
| Ownership chain | The employment, commission, assignment or other ownership route and its supporting record. |
| Third-party review | The review’s scope, findings and unresolved questions. |
| Release and customer claims | The proposed use, rights promise, exceptions and accountable review decision. |
Jurisdiction and creation basis are separate context fields. If the intended legal scope is outside the United States, the instrument marks the record outside this article’s reviewed scope. If the basis supplied is AI-only or prompts-only, it flags the need to examine any human-authorship claim. It does not calculate a percentage of copyright ownership.
DN AI Copyright Evidence Record v1.0
Seven documentation areas. No rights score or legal approval. Use reference aliases, not private documents or personal contributor details. Fields are processed in your browser; the tool makes no external requests.
No record prepared. Enter non-sensitive reference labels and build an evidence inventory.
Your evidence inventory
Reference labels, statuses and counts are your entries. The export timestamp is the time of record creation, not evidence verification.
Use aliases and reference labels. Keep confidential contracts, personal contributor details, client files and legal advice in your own controlled document system. The exported record is a pointer to evidence, with user-entered labels and a creation timestamp. That timestamp records when the file was generated; it does not verify when any source was checked.
Three Invented Scenarios That Expose Different Gaps
A generated illustration for a campaign
The team records the asset version, service terms, input history and release brief. Human contribution remains unrecorded, the ownership review is missing and a third-party review has raised an issue. The instrument reports four references, two missing areas and one explicitly unresolved area. Those counts describe the paperwork. They cannot establish whether the campaign may run or whether its image is protectable.
The useful next action is specific: ask the release reviewer to assess the actual illustration, the supplied materials and the proposed rights promise. Buying another subscription does not fill those gaps.
An edited article with generated supporting images
A writer retains drafts showing their text contributions and an editor records the finished arrangement. The images have a different creation history. The team creates separate component records so that a statement about the text is not silently extended to every pixel in the illustration.
For a registration application, the Copyright Office’s guidance requires disclosure of AI-generated content and identification of the human contribution; more-than-minimal generated material is to be excluded from the claim. Applicants should use the current official instructions for the actual work. USCO registration guidance.
A software component built with an agent
The repository contains generated code, a developer’s edits and dependencies. The team records commits and component origins, then assigns a reviewer to the relevant rights and release questions. A passing test suite supports a functional claim. It does not populate the ownership chain or dependency-permission fields.
The same record can travel with the commercial handover. It gives the customer a clearer account of what was delivered and which rights claims were actually reviewed, while preserving any explicit exceptions.
How to Run a Useful Review Without Turning It Into a Paperwork Exercise
Start with the promise. Is the business offering a usable deliverable, a limited license, exclusive rights or an outright assignment? Capture the exact proposed scope before collecting evidence. A review designed for one promise may not answer the questions raised by another.
Fix the asset version. A review attached to an older draft can become misleading after a replacement image, regenerated passage or dependency change. Use an existing file version, repository commit or internal asset identifier. Keep the identifier stable in the exported record.
Collect the minimum meaningful history. Preserve relevant inputs, substantive revisions and contribution notes. Thousands of prompts alone can obscure the decisions that matter. Select records that help a reviewer connect the actual deliverable to the claimed contribution.
Assign each gap. An unresolved field should have an owner in the team’s task system, a question and a review deadline. The public-facing tool intentionally avoids collecting personal details; internal operational assignments can remain in the team’s own systems.
Record the decision separately. The tool’s output is an evidence inventory. A reviewer’s conclusion, approved contractual wording and any legal advice belong in a distinct controlled record. This prevents a machine-readable completeness count from becoming an accidental approval badge.
Reopen after material changes. Change the record when the asset, input materials, relevant agreement or proposed use changes. A previously complete evidence inventory should not be copied blindly onto a new deliverable.
DN Alpha Thesis: The Commercial Asset Is a Reusable Rights Record
DN’s inference is that teams selling AI-assisted work can create more durable trust by making their rights claims inspectable. The advantage comes from linking an asset version, a precise promise and supporting evidence in a form the next person can reuse.
That record can support procurement, client handover and a later review. It can also reveal where the business should narrow its promise. Its value should be tested in actual work: time needed to answer a client’s rights question, missing documents at handover and revisions caused by unsupported claims. DN has not measured improvements in those outcomes for this release.
A spreadsheet or existing project system may be sufficient. Teams comparing their broader production stack can use DN’s AI content marketing platform guide. A tool purchase should follow a defined workflow need. No product is presented here as conferring copyright or providing automatic rights clearance.
Methodology, Limits and Change Log
Version 1.0, 11 November 2026. DN checked the primary sources linked below, reviewed existing DN coverage for overlap and created a seven-area evidence inventory. The instrument counts user-entered statuses, preserves unknowns and raises explicit scope or creation-basis questions. It provides no numerical legal risk model.
A complete export can contain mistaken claims or weak references. Its inputs need substantive review. This article does not evaluate a reader’s contract, establish a person’s authorship, clear third-party rights or cover every relevant jurisdiction. Obtain qualified advice where a consequential ownership or release decision depends on those facts.
Recheck after relevant legislation, appellate decisions, Copyright Office guidance or material workflow changes. Corrections should identify the affected statement and primary source: press@decentralised.news.
Commercial disclosure: This edition contains no sponsored provider placement. The evidence record is available without buying a tool. Linked DN comparison guides may contain their own disclosed affiliate links. Commercial relationships do not determine the legal-source analysis.
Frequently Asked Questions
Does a paid AI subscription guarantee copyright ownership?
No. Examine the agreement and the work separately. A payment record does not establish the human contribution, initial ownership or any necessary transfer.
Can an AI-assisted work contain protected human material?
Yes. The Copyright Office recognizes qualifying human expression in AI-assisted works, with the scope assessed case by case. USCO Part 2.
Does editing a fixed percentage guarantee protection?
The reviewed sources establish no general percentage rule. Document the actual contribution and obtain a fact-specific assessment rather than using an editing quota.
Does the DN record approve commercial use?
No. It counts documentation states and exports the entered record. It does not inspect the asset, verify references or make a legal release decision.
Can a non-US business use the worksheet?
It can organize a US copyright question, but cross-border activity may engage additional laws. Selecting another jurisdiction preserves the record while clearly marking it outside the article’s reviewed scope.
Primary Sources
- US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, January 2025.
- D.C. Circuit, Thaler v. Perlmutter, No. 23-5233, 18 March 2025.
- Supreme Court docket 25-449, including the 2 March 2026 denial of review.
- USCO registration guidance for works containing AI-generated material, 16 March 2023.
- 17 USC §102, subject matter of copyright.
- 17 USC §201, ownership of copyright.
- 17 USC §204, execution of ownership transfers.
- USCO Circular 30, works made for hire.
- USCO Circular 1, copyright basics.
About the Author
