AI companies are actively seeking creative work to train their models, and that means creators are getting offers — sometimes formal, sometimes buried in a terms-of-service update they almost skimmed past. Before you respond to any of them, it helps to know what you are actually being asked to give up.
Why Creators Are Being Approached About AI Deals Right Now
Training a large AI model requires enormous volumes of high-quality human-made content. Writers, illustrators, musicians, and photographers all have exactly what these companies need. The approaches vary: some come as direct licensing offers with a number attached, others arrive as a quiet checkbox added to a platform’s updated terms.
The problem is that many creators do not realise they have already agreed to something until the deal is done. A platform you joined years ago to share your portfolio can update its terms, send a notification email that reads like routine housekeeping, and — if you keep using the service — treat that as acceptance.
Understanding what an AI licensing agreement actually transfers is the first step before any conversation goes further.
What an AI Licensing Agreement Actually Transfers
A licence is not a sale of your copyright, but the rights it grants can be so broad that the practical difference is small. Three pairs of terms matter most:
- Exclusive vs. non-exclusive: Exclusive means only they can use it. Non-exclusive means you can still license the same work to others — but check whether the agreement says anything that limits that in practice.
- Perpetual vs. time-limited: A perpetual licence does not expire. Even if you walk away from the deal, a model trained on your work keeps using what it learned.
- Sublicensable vs. restricted: Sublicensable means the company you dealt with can pass your work to a third party — another developer, a partner, an acquirer — without asking you again.
Scope matters too. Training a model, fine-tuning it for a specific task, embedding it in a commercial product, and generating outputs that are sold to customers are all different uses. An agreement that covers all of them for a single flat fee is a very different thing from one that covers only initial training.
The Five Questions to Ask Before You Sign Anything
Before you respond to any offer, get clear answers to these:
- Who exactly receives the licence — the company you see, a parent company, or any future acquirer?
- Can they sublicense or assign your work to other AI developers or platforms?
- Does the licence survive if you terminate the agreement, and if so, for how long?
- Will your work train a general-purpose model or a specific product, and does that distinction change your compensation?
- Is there an audit right — meaning can you verify how your work was actually used and whether usage stayed within the agreed scope?
That last one is rarely offered without being asked. Without it, you have no way to know whether your work ended up somewhere you never agreed to.
Red Flags Hidden in the Standard Language
Phrases like “to improve our services” or “for product development” are intentionally vague. They can cover AI training without naming it. If you have signed a platform agreement containing that language in recent years, it is worth re-reading it.
Watch for clauses granting rights to “derivatives.” A model that learned your style and can now produce work that resembles it is arguably a derivative of your work — that argument is genuinely unsettled, which is exactly why the language matters.
Other things to flag:
- Lump-sum buyouts with no royalty structure mean you receive nothing if the model generates significant commercial value from what it learned from you.
- Waiving moral rights — a concept recognised in many countries, though only narrowly in the US — can strip your ability to object to how your work appears in outputs, separate from copyright entirely.
- Clauses allowing the company to use your name or professional reputation alongside AI-generated content are a likeness rights issue, not just a copyright issue.
What If You Have Already Agreed Without Realising
If you have been on a platform for years, there is a reasonable chance you accepted something like this already. That is not the end of the matter, and there are concrete things to check this week.
- Look for an opt-out setting. Many platforms added one after public pressure, and it is often buried in account settings under “privacy”, “data”, or “content preferences” rather than anywhere obvious. Turning it off usually applies to future training rather than past use, but it stops the clock.
- Check whether the opt-out is retroactive. Some apply only to work uploaded after you toggle it. If so, consider whether older work should come down or be reuploaded under the new setting.
- Record your objection in writing. Email the platform’s support or privacy contact stating that you do not consent to your work being used for AI training. It may not change anything immediately, but it creates a dated record of your position, which matters if the law shifts or a class action forms later.
- Read the version history if one exists. Some platforms publish dated changelogs of their terms. Finding the exact date the AI clause appeared tells you what you agreed to and when.
None of this undoes training that has already happened. It does change what happens next, and it puts you on record.
What Fair Compensation Actually Looks Like
There is no settled market rate yet for AI training licences. That is not a reason to accept the first number offered.
One-time flat fees are the current norm. Residual or usage-based models — where you receive ongoing payment tied to how the model is used commercially — exist but are rare. They are worth asking for.
Consider the context of your work within the dataset. A small curated set of a few hundred high-quality illustrations used to train a premium commercial product has different value than bulk ingestion of millions of images. If the company is being selective, your individual contribution is worth more.
Some creator guilds and collectives are negotiating group rates, which gives individual creators a real benchmark. If compensation is low, ask whether attribution, opt-out rights for future products, or a right of first refusal on follow-on deals can be added instead of more money.
Protecting What You Keep After the Deal
A licence should not transfer ownership of your copyright, but sloppy drafting sometimes does. Get it confirmed explicitly in writing: you retain ownership, and the licence is limited to the agreed scope.
Specific things to address before signing:
- Reserve the right to keep publishing, selling, and licensing your work independently — unless exclusivity is being paid for generously.
- Define what happens to your work if the company is acquired, merges, or goes bankrupt. Licences are assets that transfer in those events.
- Ask for a clause requiring deletion of your work from training datasets if you terminate — even if deletion from the trained model itself is technically not possible, this documents your intent and may matter later.
- Keep a record of every version of every work you licensed, when, and to whom. Scope violations are hard to prove if you cannot show what was agreed.
When to Walk Away and What to Do Instead
If a company refuses to clarify vague terms or pressures you to sign quickly, treat that as information about how they will behave after you sign.
Opt-out registries and signals like C2PA provenance metadata or robots.txt directives are not legally binding in most jurisdictions — a lawyer in your country can confirm what applies to you — but they establish a documented position and create a record of your intent.
Some creators are choosing to license selectively to smaller, transparent AI projects where the use case is specific and auditable. That approach trades scale for clarity.
If you are unsure whether an offer is reasonable, a single hour with an intellectual property lawyer costs far less than a bad perpetual sublicensable deal. Saying no is a valid business decision. Your back catalogue is a long-term asset, and the market for these licences is still forming.
Frequently asked questions
Can I license my work for AI training and still sell or publish it myself?
Generally, yes — if the licence is non-exclusive. A non-exclusive licence lets you continue using and licensing your work to others. An exclusive licence would restrict that. Read the agreement carefully, and if the language is unclear, ask the company to confirm in writing that you retain independent publishing and licensing rights.
If I agree to an AI licence now, can I change my mind later?
It depends on what the agreement says. Many AI licensing agreements include termination clauses, but they also often state that the licence survives termination for work already incorporated into a trained model. This varies by contract and by country. Before signing, ask specifically what happens to your work if you terminate — and get the answer in writing.
Do platform terms of service count as an AI licensing agreement for my uploaded work?
They can. Platform terms of service are contracts, and if they include language granting rights to use uploaded content for “product improvement” or “service development,” that language may cover AI training. This varies by platform and by jurisdiction. If you are concerned about a specific platform, re-read its current terms and consult a lawyer if the language is unclear.
I think I already agreed years ago. Is it too late to do anything?
Not entirely. Check your account settings for an AI or data-use opt-out — many platforms added one and did not advertise it. Find out whether it applies retroactively or only to new uploads. Then put your objection in writing to the platform, so there is a dated record of your position. None of this reverses training that has already taken place, but it governs what happens from here and may matter if the legal position shifts.
Everyone deserves authenticity and accountability online. That is the mission we are working on. Subscribe to join us, and get a free copy of Wes Kussmaul’s Escape the Plantation.
This article is general information, not legal advice. Laws differ by country — consult a qualified lawyer about your situation.
