Before You Licence Your Work for AI: Key Terms to Check

Define the AI uses you permit and keep dated evidence of the versions you provide.

An AI company may ask to use your writing, images, music or other creative work. Before handing over files, make sure the proposed agreement spells out what the recipient can do with them, who else may get access and what happens when the arrangement ends.

Treat “AI training” as a starting point for discussion, not a complete description of the permission being requested. Check the scope, duration, payment and records together, and keep evidence of the specific work and version you agree to provide.

Start by identifying the material

Ask for an inventory attached to the agreement. For each item, it could record the title, filename, format, version, author, rights holder and any deposit reference. Avoid relying on broad phrases such as “your content” or “your catalogue.”

If you are considering a book, clarify whether the request covers only the manuscript or also its illustrations, photographs, translation and cover art. For a recording, identify the particular components included in the proposed permission.

Ask how the parties will handle drafts, unpublished files, metadata and future revisions. Can a replacement upload add new material to the deal without your written approval? Set out exclusions and a process for adding items later.

Record the work and version before delivery

Keep a dated record of the original works and exact versions named in the agreement. A certificate from an online deposit can support evidence of authorship and show that the deposited version existed by the recorded date. It does not create copyright.

Keep the certificate with the signed agreement, inventory, delivery confirmation, approvals and any later amendments. Cross-reference deposit identifiers in the inventory, and retain drafts and source files that help document your creative process.

The deposit and the agreement serve different purposes: one records a version of the work, while the other sets out the permission being granted. Keeping the records together can help clarify what was supplied and which terms applied to it.

Canadian copyright and your AI agreement

In Canada, copyright protection for an original work starts automatically when the work is created; registration is not required to create the right. Copyright protects the way an idea is expressed, not the idea itself. A private deposit does not create copyright, but a dated certificate can help show authorship and when the deposited work existed.

A licence should identify the work and the uses you are allowing. If a question arises about a particular version or date, proof of authorship and the date can matter. Registration with the Canadian Intellectual Property Office is optional; a CIPO certificate is evidence that copyright exists and that the registered person owns it.

In Canada, an author’s moral rights cannot be assigned, though they can be waived in whole or in part. Check the agreement for any waiver language and consider getting advice before signing. Copyright generally lasts for the author’s life and until the end of the 70th year after their death; the term change did not bring back works by authors who died before 1972.

Spell out every authorised AI activity

Ask the other party to distinguish among pre-training, fine-tuning and retrieval-augmented generation (RAG). With RAG, a model connects to an external knowledge base to draw on information when responding. Make sure a permission to train does not quietly include a separate permission to make your work available for retrieval.

Ask whether the agreement covers copying and preparing files, training identified models, fine-tuning, performance testing, retrieval, synthetic training material or development of other models. Have the permitted activities listed rather than relying on an undefined reference to AI use.

Consider public-facing uses separately. Could the recipient show excerpts, reproduce images, offer downloads or feature your work in demonstrations or advertising? If you want those activities to require separate approval, say so in the agreement.

You can also ask about safeguards, attribution and a process for raising concerns about outputs. Keep any negotiated limits focused on the permission you are granting; do not treat them as a claim that copyright protects a style, subject or idea in general.

Specify who can access the files

Identify the legal entity signing the agreement and ask which developers, affiliates, contractors or customers may receive access. Clarify whether subcontractors may handle the material only for the named recipient or can obtain permission to reuse it independently.

Discuss sublicensing, resale of datasets, corporate transfers and distribution of model weights. Does each require your consent, notice or another agreement? Identify the territories where storage, training and commercial use are permitted.

You could ask the agreement to name approved recipients and require equivalent restrictions for downstream users. Consider asking a qualified adviser to review the governing law, dispute process, warranties and liability before you sign.

Set clear start, end and exit terms

Use distinct time periods for file access, preparation, training, storage and operation of trained models. A short training period does not necessarily mean that use of the resulting model also ends quickly, so ask how long each permission lasts.

At expiry or termination, what must happen to source files, prepared datasets, backups and copies held by contractors? Must further training stop? Can an already trained model remain in use, and under what conditions?

Replace vague promises to “delete the data” with specific terms about deletion, trained models, verification and deadlines. Also clarify how renewal, non-payment or a serious breach affects each permission.

Make the payment terms measurable

The proposed fee might be a one-time amount, recurring royalties or a combination. Ask how the payment structure relates to the scope and duration of the permission, rather than assuming one payment covers every use that might arise later.

Identify when payment is due: on signing, delivery, acceptance, ingestion, training or commercial release. Ask what happens if the recipient accepts the files but never releases a product.

For payments based on use, request a calculation you can check. Define what counts as a work, training run, model or revenue event, and clarify any minimum guarantee, deductions, currency, taxes, payment dates and corrections.

Would additional models, new activities, longer retention or expanded territories require fresh approval and payment? Put the answer in writing.

Ask for records that can be checked

Request reports that connect to the agreement’s commitments. Depending on the deal, those reports might identify the works received, versions ingested, datasets and models involved, training dates, authorised recipients and payment calculations.

Agree on how often reports will arrive, what format they will use and who is responsible for them. Ask what records the recipient will retain, how long they will keep them and how you can question incomplete information.

You can propose an independent audit of relevant records, subject to confidentiality protections. Discuss who pays and how errors will be corrected. If confidential information cannot be disclosed, ask what other method could verify compliance. These are possible contract terms, not reporting rights that every creator automatically has.

Check ownership and authority to grant permission

Before promising rights over every item in a file, check who created and controls the material. Employment arrangements, earlier agreements, co-authors and third-party contributions may affect who can authorise its use.

A licence generally describes permission to use specified rights, while an assignment may transfer rights. The legal effect depends on the agreement and applicable law. Ask which rights you keep, whether the permission is exclusive and whether it limits your ability to licence the work to others. Moral rights and other contractual requirements may also need attention under the applicable law.

Questions to settle before you agree

• Are the works, versions and exclusions clearly listed?
• Are the AI activities, models, recipients and territories specified?
• Do the agreement’s duration, renewal and exit terms cover both files and trained models?
• Can you check when payment is due and how it is calculated?
• Are reporting and verification arrangements set out?
• Are the dated records for the identified works kept with the signed terms?

Protect a dated record of your work

Use Scopyright.ca to deposit your work online and receive a dated certificate that supports evidence of authorship and the date of the deposit. Copyright exists from creation; the certificate helps you document the work and version you may be licensing.

Start with the copyright protection options for Canada and keep your certificate with the agreement and delivery records.

Make the agreement fit the permission

Decide which uses you are willing to allow, then ask for wording that matches that decision. A clearly scoped agreement and an organised record of the identified work can help both parties understand what was provided, what was permitted and what was left out.

Sources

• WIPO — Summary of the Berne Convention for the Protection of Literary and Artistic Works (1886)
• WIPO — What Can I Protect with a Copyright?
• WIPO — IP Assignment and Licensing
• WIPO — Frequently Asked Questions: Copyright
• UK Government — Report on Copyright and Artificial Intelligence
• WIPO — Understanding Copyright and Related Rights

Article written with the assistance of AI. General information, not legal advice.