TREN
Guide · Copyright

Is what you generated actually yours?

Once AI-generated images, text and code become part of a commercial product, two separate questions arise: can you use it, and can you stop anyone else from using it? The answers are often different. Below is the checklist that produces a decision.

Generative AI output copyright and commercial use
Download the checklist Reserve your spot for an online call
§ 01 — Two separate questions

"Can I use it" and "do I own it" are not the same question.

Marketing generates the campaign visual with AI; engineering has the assistant write part of the code. Once the product ships, the question asked is usually a single one: "could this come back at us?" But two distinct issues are being run together here.

First, permission to use. This looks primarily to the provider's terms, and in most tools it varies by subscription tier. Output generated on an individual account and used in a commercial product is the most common mismatch in practice.

Second, ownership. Copyright protection generally requires a work that is its author's own intellectual creation, with authorship attributed to a natural person. Whether purely machine-generated output with no human input qualifies is contested.

The practical consequence

If the output is not a protected work, that does not stop you using it. It does mean you may be unable to stop a competitor using the same thing. For a logo or a hero image that will carry your brand identity, that is a commercially significant difference.

This is why the "record of human contribution" section of the checklist is not a formality: documenting who made the creative choices, which alternatives the output was chosen from and on what criteria, and what was edited afterwards, strengthens any claim you later need to assert.

With code, the risk is licensing, not copyright

Code output behaves differently. The risk there is that the output reproduces a copyleft-licensed fragment and that licence's obligations propagate across your product. Run a licence scan before commercial use and record any detected components in your inventory (SBOM).

§ 02 — Legal basis

What is engaged.

BasisSubjectWhat it means in practice
Copyright lawOriginality"Author's own intellectual creation" threshold
Copyright lawAuthorshipAttributed to a natural person
Employment termsWorks by employeesTransfer of rights to the employer
AI Act Art. 50(2)MarkingMachine-readable marking of synthetic content
AI Act Art. 50(4)Deepfake disclosureDepicting a person doing what they did not do
AI Act Art. 53GPAI providersCopyright policy and training-data summary
Dir. 2019/790 Art. 4Text and data miningRightholder's machine-readable reservation
Licence termsCode componentsPropagation of copyleft obligations

Provisions reflect the text as at the date this page was prepared. Authorship analysis differs between jurisdictions — confirm the position in each market you sell into.

§ 03 — Checklist

Copyright and commercial-use checklist.

Complete a separate one for each family of output. Copy it or download it as markdown. No sign-up.

AIA-IPR-04 · Version 1.0 · Free Download .md ↓
# GENERATIVE AI OUTPUT — COPYRIGHT AND COMMERCIAL USE CHECKLIST

Document code: AIA-IPR-04 · Version 1.0 · Classification: Internal
Assessment: ……/……/20……   Renewal: on each commercial use
Complete a SEPARATE checklist per family of output.

PART A — OUTPUT RECORD
A.1 Output type: ( ) Image ( ) Text ( ) Code ( ) Audio/video ( ) Other
A.2 Tool and version used: [……]
A.3 Subscription tier: ( ) Individual ( ) Enterprise ( ) API
A.4 Date generated: [……]
A.5 Intended commercial use: [……]
A.6 Territories of use: [……]
A.7 Prompt record retained: ( ) Yes ( ) No

PART B — OWNERSHIP
B.1 Protection generally requires a work that is its author's own
    intellectual creation, with authorship attributed to a NATURAL PERSON.
    Whether purely machine-generated output qualifies is contested.
B.2 PRACTICAL CONSEQUENCE: if the output is not a protected work you may be
    unable to assert exclusive rights. You can use it; you may not be able
    to stop others using the same thing.
B.3 RECORD OF HUMAN CONTRIBUTION (strengthens any claim):
    B.3.A Person making the creative choices and direction: [……]
    B.3.B Edits and adaptations applied: [……]
    B.3.C Which alternatives it was chosen from, on what criteria: [……]
    B.3.D Is this record retained: ( ) Yes ( ) No
B.4 Where the contribution was by an employee, check the contractual
    provision transferring rights to the employer.

PART C — PROVIDER TERMS
C.1.A Is commercial use of the output permitted?               ( ) Y ( ) N
C.1.B Does commercial use depend on subscription tier?         ( ) Y ( ) N
C.1.C Does the provider assert rights over the output?         ( ) Y ( ) N
C.1.D Are inputs used for model training?                      ( ) Y ( ) N
C.1.E Is there an indemnity for third-party claims?            ( ) Y ( ) N
C.1.F On what conditions: [……]
C.2 If C.1.B is "Yes": verify the generating account was ACTUALLY on that
    tier. Individual-account output used commercially is the most common
    mismatch.

PART D — INFRINGEMENT SCREEN
D.1 IMAGE
    D.1.A Recognisable trade mark, logo or trade dress?         ( ) Y ( ) N
    D.1.B Recognisable image of a living person?                ( ) Y ( ) N
    D.1.C Directly imitates a specific artist's style?          ( ) Y ( ) N
    D.1.D Reverse image search performed?                       ( ) Y ( ) N
D.2 TEXT
    D.2.A Verbatim passages from source texts?                  ( ) Y ( ) N
    D.2.B Plagiarism scan performed?                            ( ) Y ( ) N
    D.2.C Unverified factual assertions?                        ( ) Y ( ) N
D.3 CODE
    D.3.A Licence compatibility scan performed?                 ( ) Y ( ) N
    D.3.B Copyleft (GPL etc.) fragments detected?               ( ) Y ( ) N
    D.3.C Licence obligations for detected components met?      ( ) Y ( ) N
    D.3.D Recorded in the component inventory (SBOM)?           ( ) Y ( ) N
D.4 For every "Yes" in D.1 or D.3, a written assessment is made BEFORE use.
    Not having run the scan weakens any good-faith defence.

PART E — MARKING AND TRANSPARENCY
E.1 Machine-readable marking of public synthetic content (Art. 50(2)):
    ( ) Applies ( ) Does not apply
E.2 Deepfake disclosure (Art. 50(4)):   ( ) Applies ( ) Does not apply
E.3 Public-interest AI text disclosure: ( ) Applies ( ) Does not apply
E.4 GPAI providers' copyright policy and training-data summary duties
    (Art. 53) are reserved; ask about them in supplier selection.

PART F — TRAINING DATA (if you train or fine-tune)
F.1.A Source of the training data and how obtained: [……]
F.1.B Machine-readable reservation (opt-out) checked?          ( ) Y ( ) N
F.1.C Does the data contain personal data? ( ) Y ( ) N (GDPR assessment)
F.1.D Terms of licensed data sets recorded?                    ( ) Y ( ) N

PART G — DECISION
G.1 ( ) Cleared for commercial use
    ( ) Cleared subject to conditions: [……]
    ( ) Not cleared — reasons: [……]
G.2 Reasoning (mandatory): [……]
G.3 Retained with reasoning and date. It is the first document requested
    when a copyright claim arrives.

SIGNATURE: Assessed by / Legal review / Approved by

This checklist is general in nature and does not constitute legal advice.
§ 04 — Filling it in

The three parts most often skipped.

  1. Part C.2 — tier verification. The provider terms get read, but nobody checks which tier the generating account was actually on. If commercial permission is tied to the enterprise tier, an image made on an individual account does not benefit from it.
  2. Part B.3 — record of human contribution. This record cannot be reconstructed later. If it is not kept while the choices and edits are being made, there is no document to support a claim.
  3. Part D.3 — the code scan. Code output never reaches this checklist in most companies, yet propagation of copyleft obligations into a product is the commercially costliest scenario.
Note This checklist is a general framework and does not substitute for an ownership analysis. Authorship and infringement turn on the specific output, the tool used and the markets you sell into; on the code side, licence obligations produce different outcomes depending on your product architecture. You can book a preliminary call to adapt the checklist to your output and review the indemnity in your provider agreement.
§ 05 — Frequently asked

Questions.

Can I use an AI-generated image commercially?

Permission depends first on the provider's terms and usually varies by subscription tier. Even where permission exists, check separately whether the output reproduces a protected third-party work.

Who owns the output?

Protection generally requires a work that is its author's own intellectual creation, with authorship attributed to a natural person; purely machine-generated output is contested. The result: you can use it, but you may not be able to stop others from doing the same.

Can I ship AI-generated code?

The principal risk is licence compatibility rather than copyright. If the output reproduces a copyleft fragment, those obligations may propagate across your product. Scan before commercial use and record components in your inventory.

Do I have to mark the content?

Art. 50(2) requires machine-readable marking of synthetic content made available to the public. Where content depicts a person doing something they did not do, Art. 50(4) requires disclosure as well.

Does the provider's indemnity protect me?

Partly. These undertakings are usually conditional — a particular tier, filters left enabled, output left unmodified. Read what the indemnity is conditioned on before relying on it; the checklist asks for exactly that at C.1.F.

Related

Read next.

Let's assess your output together.

In a twenty-minute preliminary call we look at which families of output you generate and work out which carry risk and what your provider terms actually say. If your exposure is low, we tell you that clearly too.

Reserve your spot for an online call The call is online · the checklist is yours either way