AI Trends
Trump’s AI Copyright Position: What It Means for Creators and Training Data
Understand Trump’s AI copyright position, the courts’ role in fair use, and the questions creators should ask about training-data licenses.
Trump’s AI copyright position supports the use of copyrighted material in model training, while leaving the fair-use dispute to the courts. The administration’s March 2026 framework also proposes ways for creators to negotiate licensing payments. These are different questions: whether permission is legally required, and how a licensing arrangement might work.
The White House’s March 20, 2026 announcement presents the framework as recommendations for Congress to turn into legislation. Read it as a policy proposal, rather than a court judgment or a license to use someone’s work.
This guide examines that framework alongside the Copyright Office’s published analyses. It is a U.S. policy explainer, not advice on whether a particular dataset, contract, or output is lawful.
What the Trump framework says about copyright
Section III of the White House’s legislative recommendations combines three positions:
- Training: The administration believes training on copyrighted material does not violate copyright law, acknowledges opposing arguments, and supports judicial resolution. It asks Congress to avoid interfering with that determination.
- Licensing: Congress should consider arrangements allowing rights holders to negotiate collectively with AI providers without antitrust liability. The proposed legislation would not determine when licensing is required.
- Creator protection: The framework supports protection against infringing outputs and proposes safeguards for unauthorized digital replicas, with exceptions for protected expression.
The licensing proposal therefore does not promise an automatic payment to every creator whose work appears in a dataset. Nor does support for training settle a dispute over a particular output. When assessing a headline, identify which of these questions it actually addresses.
Why fair use still requires a factual analysis
The Copyright Office’s explanation of fair use describes four factors courts weigh together. There is no universally safe percentage of a work that can be copied.
| Factor | Question for an AI training dispute |
|---|---|
| Purpose and character | What is the use for, how does it differ from the original purpose, and is it commercial? |
| Nature of the work | Is the material factual or highly creative, published or unpublished? |
| Amount and substantiality | How much was copied, and how significant was that material within the work? |
| Market effect | How might the use affect the original work’s existing or potential market? |
The questions in the table apply the Office’s general framework to the training context; they do not predict a ruling. Calling a use “transformative,” “educational,” or “commercial” does not complete the analysis.
How the Copyright Office’s training report compares
The Office’s May 2025 Part 3 report on generative AI training, in its pre-publication version, takes a fact-dependent approach. Its conclusion recognizes that some training uses are likely transformative, but emphasizes the source of the works, the purpose, output controls, and market effects.
The report distinguishes analytical or research uses from commercial systems producing expressive content that competes with training works. It is particularly critical where the latter involves illegal access. It also favors continued development of voluntary licensing, with targeted alternatives considered if gaps persist. These are the Office’s analysis and recommendations, not a judgment deciding every training case.
The editorial distinction is that the White House states a broad policy preference, while the Office examines circumstances that could change the legal result. Neither document supplies the missing facts about a specific developer’s dataset. Before drawing a conclusion, ask what was acquired, how it was acquired, and what the system does with it.
Three creator situations that need different evidence
The following examples are hypothetical. Use them to organize questions before making a claim or accepting an offer.
A writer receives an offer to license an archive
Ask for an itemized list of the articles, editions, and translations covered. Mark which pieces you wrote alone and which involve a publisher or collaborator. Request an explanation of permitted uses and payment calculations. Keep the proposed agreement alongside the contracts under which the archive was originally published.
If you also use our writing prompts to develop new work, keep that project in a separate folder. A proposed license for your back catalogue deserves its own review, independent of your drafting habits.
A photographer finds a suspiciously similar generated image
Preserve the original photograph, the image you encountered, its URL, and the date. If you generated the comparison yourself, record the exact prompt, tool, and relevant settings. Describe observable similarities precisely; do not treat visual resemblance alone as a complete account of how a model was trained.
For experiments such as our photo-to-watercolor illustration prompts, start with a photograph you are authorized to use and retain it with the result. That creates a clearer record of what you supplied and what changed.
A publisher evaluates an AI partnership
Request separate descriptions of archive access, training, search or retrieval, and display of excerpts. Ask which content is included, who receives it, and whether another company may reuse it. Build a list of unanswered questions before comparing the headline payment with another offer.
For a smaller commercial project using our product-photo advertising concept prompts, use the same documentation habit: list the source assets, the proposed transformations, and the intended publication channels in the project brief.
What to ask before accepting an AI training license
Use this editorial checklist to prepare for a contract review. It identifies information to request; it does not establish which rights you own or which terms are enforceable.
- Coverage: Which exact works, file versions, and future additions are included?
- Purpose: Does the agreement distinguish training, evaluation, retrieval, and output display?
- Recipients: Which named companies may access the material, and is sublicensing contemplated?
- Payment: What triggers payment, how is it calculated, and what records can you inspect?
- Duration: What happens at expiry or termination, including to copies and previously trained models?
- Output handling: What process addresses reproduction of supplied content or a complaint about an output?
- Representation: If a collective negotiates, how are membership, authority, fees, and distributions explained?
For example, “access for three years” leaves a practical question unanswered: what use, if any, continues after access ends? Ask for an explicit written answer rather than filling the gap with an assumption.
A prompt for organizing the review
Use this with text you are permitted to share with the chosen tool. Check its output against the document yourself.
Review the proposed AI content license pasted below.
Extract only what the text expressly says about:
1. Works covered and exclusions
2. Training, evaluation, retrieval, and display rights
3. Recipients and sublicensing
4. Payment and reporting
5. Duration and post-termination use
6. Output complaints and remedies
Return a table with: issue, exact clause, plain-language summary,
and question to ask the other party.
Write "Not stated" where information is missing.
Flag ambiguous wording without inventing its meaning.
Do not infer ownership, decide enforceability, or declare the deal safe.
Agreement text: [paste authorized text]Training rights and output ownership are separate questions
Whether a developer could lawfully train on an image does not answer whether a user can claim copyright in a generated illustration. The Copyright Office’s January 2025 Part 2 report on copyrightability centers the latter question on human authorship. It recognizes protection for qualifying human expression within works involving AI, while excluding purely AI-generated material. It also concludes that prompts alone, under the technology it examined, do not supply sufficient expressive control.
Keep your original photographs, drafts, selections, and substantive edits together. That record makes your contribution easier to explain; it does not guarantee a particular copyright determination.
How to assess the next Trump AI copyright headline
Open the underlying document and record its date. Identify whether it is a policy statement, a bill, an enacted law, a filed argument, a judgment, or a private agreement. Then write down the precise activity it concerns: acquiring training material, training a model, licensing content, producing an output, or claiming authorship.
Before changing your own workflow, ask one final question: which documented fact or applicable rule changed for this project? Save that answer with the source. It will be more useful than a headline when you next discuss permissions, payment, or publication with a collaborator.
