iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Sometimes, but there is no industry-wide yes or no. A game-related model may be trained on creator uploads, company-owned material, commercially licensed datasets, open-license or public-domain works, third-party data obtained under contract, or synthetic examples. The answer depends on the named model, its task, the developer’s disclosed sources, the rights attached to each item and the law in the relevant country.
Roblox says some creator content can be included when creators have not opted out and warns that training data may carry third-party intellectual-property rights. KRAFTON says its inZOI systems used separately selected sources it considered commercially permitted, including open-license images and company-owned or commercially permitted 3D and motion data. Those are disclosures about particular systems, not audits of every game’s training corpus.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Copyright Law | $140.00 | Buy on Amazon |
| 2 |
|
Copyright Law: Cases and Materials (v8.0) | $21.70 | Buy on Amazon |
| 3 |
|
Copyright Law of the United States: and Related Laws Contained in Title 17 of the United States Code | $10.32 | Buy on Amazon |
| 4 |
|
Copyright Law in a Nutshell | $65.00 | Buy on Amazon |
| 5 |
|
Copyright Handbook, The: What Every Writer Needs to Know | $37.97 | Buy on Amazon |
What does “trained on copyrighted work” actually mean?
The phrase combines several questions that must be separated. A work can be protected by copyright, appear in a dataset, be used under a license, and still raise a different legal question about whether the particular copying or training use infringed. Copyright status alone does not decide infringement.
Recommended Free Tools
| Question | What it establishes | What it does not establish |
|---|---|---|
| Was the work included in training data? | Evidence that a model’s dataset contained the item or a copy of it. | Whether the item was protected, licensed, or used lawfully. |
| Was the work copyrighted? | The item may have exclusive rights attached to it, depending on jurisdiction, term and authorship. | That every use of it was unauthorized or infringing. |
| Was there permission? | A license, ownership claim, public-domain status or contract may provide a rights basis. | That the license covered every use, every territory, or every component of the dataset. |
| Was the use infringing? | A legal conclusion based on applicable law and facts. | A conclusion that can be applied identically to every country or model. |
For a useful answer, identify the model and task first: texture generation, code assistance, image generation, 3D asset creation and motion generation can all use different datasets. Then ask what the developer says about sources, rights, creator controls and the jurisdiction whose law matters.
#1 Best Overall
What has Roblox disclosed about its generative-AI training data?
Roblox’s current Creator Hub documentation describes several source categories for its public generative-AI systems, while stressing that sources vary by system. The documentation states: “Roblox’s generative AI systems made available to the public are trained on the following, with particular data sources varying for each system:”
- Publicly accessible permissively licensed or public-domain repositories.
- Non-public data obtained from third-party providers through commercial arrangements.
- Platform activity and creator material when users have not opted out.
- Internally generated material, including synthetic data.
- Material from annotation services and contractors.
The listed creator-facing tools include Code Assist, Material Generator, Assistant, Texture Generator and Avatar Setup. Roblox says training data can include text and audio text, visual content, annotations, ratings and preference data. Depending on system type and version, training can range up to over 3 million distinct assets; that is a maximum-range statement, not a count for every model. The company also warns that some training material may be subject to third-party intellectual-property rights. See the Roblox AI data-sharing documentation for the current policy language.
What are Roblox’s sharing defaults and withdrawal timelines?
Roblox says sharing is off by default for games, avatar items and paid assets published before July 10, 2024, and on by default for those published on or after July 10, 2024. These defaults are specific to Roblox and can change, so the publication date matters when interpreting a creator’s control.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
If a creator stops sharing, Roblox says it removes the content from the training dataset within 30 days and updates models trained using that data within 365 days. Removing an item from a dataset and updating a model are different stages; the stated timelines do not mean that every prior output disappears from circulation.
What does the Roblox disclosure prove?
It is a source-category and control-policy disclosure, not a work-by-work corpus. It does not independently verify ownership, license scope or the status of every asset. The precise conclusion supported by the document is that some creator data may be used and may carry third-party IP rights, not that all Roblox models were trained on infringing material.
What has KRAFTON said about the inZOI models?
In a May 13, 2025 notice, KRAFTON described three separate inZOI technologies and gave different accounts of their data. The company says these sources were selected as permitted for commercial use, but the statements are KRAFTON’s own descriptions rather than an independent audit.
| inZOI technology | Stated training material | Rights basis described by KRAFTON |
|---|---|---|
| Text-to-image (T2I) | 20 million publicly available images | Images released under Creative Commons licenses that permit commercial use and modification |
| 3D-printing AI | Approximately 46,000 high-quality 3D models | Models owned by KRAFTON or sourced from datasets the company says were permitted for commercial use |
| Video-to-motion | Over 1.7 million motion-data samples | Primarily data owned and collected by KRAFTON, supplemented by a limited number of publicly available open-license datasets |
KRAFTON’s notice says, “The Text to Image (T2I) technology applied in inZOI was trained using publicly available images that are permitted for commercial use.” Read the full KRAFTON inZOI notice for its descriptions of all three systems.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsA commercial-use label is not a universal guarantee. Creative Commons licenses differ, may include conditions such as attribution or share-alike, and do not automatically settle questions about third-party rights embedded in a work. KRAFTON’s figures therefore show a contrasting disclosure approach to Roblox: named quantities and a stated rights basis for particular tasks, not proof about every model or every underlying file.
Rank #3
How can you tell what a game’s AI was trained on?
Use the game’s own documentation as a starting point, but treat a feature disclosure and a training-corpus inventory as different things.
- Name the feature and task. Determine whether the game uses AI for development-time assets, runtime generation, code, dialogue, textures, 3D objects or motion. A single game can use multiple models.
- Find the developer’s model-specific statement. Look for named datasets, source categories, ownership claims, commercial licenses, public-domain material, synthetic data and third-party contracts. “AI-generated” by itself identifies neither a model nor its training sources.
- Check creator controls and dates. Defaults may differ by publication date, account type or asset category. Look for opt-out methods, deletion windows and model-update commitments.
- Separate storefront disclosure from provenance. Steam’s developer policy distinguishes pre-generated AI-assisted material made during development from live-generated content. A label can explain how AI appears in a game without listing the model’s complete training data.
- Apply the right jurisdiction. A rights statement made for one country or region may not answer the legal question elsewhere. Record the date of the policy and the territory it addresses.
This approach can establish what a developer says it used. It usually cannot prove that every item in a large dataset was correctly licensed, because public disclosures rarely provide a complete, independently verified inventory.
Does using copyrighted material for training automatically infringe?
No. Whether a particular use infringes depends on the jurisdiction, the source and the applicable legal rules, along with facts such as copying, licensing, exceptions and the purpose of the use. “Copyrighted” and “infringing” are related terms, not synonyms.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →What current policy sources say
The UK government’s Report on Copyright and Artificial Intelligence, published March 18, 2026, says generative-AI models are often trained on creative works protected by copyright and that public disclosure by developers is limited. It describes differing national approaches and notes that UK law does not currently generally require AI developers to publicly disclose the copyright works used to develop a model. The same report describes European Union AI Act requirements for providers of general-purpose AI models to publish sufficiently detailed summaries of training material. Those observations are region-specific and should not be treated as one worldwide rule.
Rank #4
The U.S. Copyright Office’s AI study includes a May 9, 2025 pre-publication Part 3 report on generative-AI training. Its January 29, 2025 announcement about Part 2 addresses a different issue: when an output can receive copyright protection. The Office said, “The Office confirms that the use of AI to assist in the process of creation or the inclusion of AI-generated material in a larger human-generated work does not bar copyrightability.” That concerns human authorship in outputs, not whether training data was licensed or whether training was infringing. The Office reports receiving over 10,000 comments by the December 2023 deadline for its AI study. See the U.S. Copyright Office AI study index and its Part 2 announcement.
Why game-specific context matters
Video Games Europe’s April 2024 position paper distinguishes earlier machine-learning uses of telemetry and other game data owned by a company from generative AI trained on creative assets, where copyright questions arise. That is an industry association’s position, not a neutral audit of every studio. It is useful context because “game AI” covers both conventional data-driven systems and newer generative models.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What should players and creators do with an unclear disclosure?
For players
- Look for a dated developer notice naming the feature, model task and source categories.
- Do not assume a Steam AI label lists the training corpus or proves that all training material was licensed.
- If provenance matters to you, ask the developer whether the system uses user-created assets, what opt-out controls exist and which jurisdiction’s policy applies.
- Treat claims such as “commercially permitted” as the developer’s stated rights basis, not as an independent legal ruling.
For creators supplying assets
- Check whether sharing is opt-in or opt-out for the asset type and publication date.
- Read how withdrawal, dataset removal and model updates are timed; those are separate commitments.
- Do not upload third-party material unless you have the rights needed for the platform’s stated uses.
- Keep a record of the platform policy version and your sharing choice, because defaults and documentation can change.
What is still unknown about game-AI training?
No reviewed source supplies a complete, independently verified inventory of training works across the games industry. Roblox identifies broad categories and a possible scale of more than 3 million assets, while KRAFTON gives model-specific quantities and rights descriptions for inZOI. Neither disclosure establishes what every other studio used.
Free tools Windows power users keep installed
One-click scans. No signup required.
Even where a developer names a license, important details can remain unresolved: whether the license covered all dataset components, whether contributors had authority to grant it, how derivatives were handled, and which country’s law governs a dispute. A careful answer therefore uses precise language: a developer disclosed certain sources; material may be copyright-protected; and the lawfulness of the use is disputed or context-dependent.
Best Value
Frequently Asked Questions
Can a game call itself AI-powered without naming its training data?
Yes. Storefront and marketing disclosures can describe an AI feature without identifying the model, dataset or rights basis. Ask for a model-specific provenance statement if those details affect your decision.
Does a Creative Commons label guarantee that training was lawful?
No. Creative Commons licenses have different conditions, and a license may not cover rights held by someone other than the uploader. The relevant license version, attribution terms, dataset selection and jurisdiction still matter.
Are AI-generated game assets automatically protected by copyright in the United States?
No. The U.S. Copyright Office says AI assistance or AI material inside a larger human-created work does not automatically bar protection, but copyright depends on a human author determining sufficient expressive elements. That output analysis is separate from the legality of training.
What evidence would make a developer’s training claim stronger?
A stronger disclosure would identify each model and task, name or characterize datasets, explain ownership or license scope, provide creator controls and dates, and describe how withdrawal affects datasets and model updates. Independent verification would provide more assurance than a statement alone.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
