Licensing a video game for AI training: the contract checklist

Daniël Paaij · Updated 15 October 2026 · 6 min read

A licence for AI training should define exactly what the lab may do with your game, where it runs, who owns what comes out of it and how you get paid. Use this checklist to prepare for talks with a lab and your lawyer. It is a practical guide, not legal advice.

In short · What belongs in a licence for AI training?

Six things: the licensed use (training, evaluation, or both, named separately), where the build may run and that no copy leaves it, who owns the models and trajectories that come out, confidentiality and the operator's separation duties, the fees and the studio's share of each, and the term with audit and exit rights. Check your own rights first — licensed music, third-party engines and middleware carry terms that can differ for server use. This is a practical checklist, not legal advice.

1. Licensed use

Purpose: training, evaluation, or both. Name them separately.
Which titles, versions and content packs are included.
Whether the lab may modify the game (maps, rules, assets) and who owns those changes.
Explicitly excluded: public release, redistribution, sublicensing and use in commercial games.

2. Where the game runs

Only on named, controlled hardware (for example the operator's dedicated environment for that lab).
Whether running inside the lab's own infrastructure is allowed, and under which security conditions.
No copies of the build outside the agreed environment; deletion at the end of the term.

3. Outputs and model rights

The lab owns its agents, models and trajectories. State this; it is what labs need to sign.
Whether models trained on your game may generate content that resembles your game's assets or characters. Most studios restrict this.
Whether the lab may publish research results, videos or benchmarks that show your game, and with which attribution.

4. Confidentiality and separation

Mutual NDA covering your build and the lab's research.
Commitment that the operator keeps each lab's environment physically separate and stores no agent data.
Access logging and the lab's right to revoke operator access.

5. Fees and revenue share

Integration fee: amount, what it covers, payment milestones.
Monthly service fee and usage fee per concurrent agent, with how usage is measured.
Your share of each, if an operator such as benned delivers the service.
Price review at renewal and minimum commitment, if any.

6. Term, audit and exit

Term and renewal, plus exclusivity if the lab asks for it (and what it pays for that).
Your right to audit usage and where the build runs.
Termination for breach, and what happens to environments and copies at the end.
Liability caps and governing law.

Before you sign: rights you need yourself

Check that you hold the rights to license the game for this use. Licensed music, third-party engines, middleware, brands and outsourced art may have their own terms. A black-box set-up and an engine integration can have different implications under your engine licence.

Preparing a first deal? We can share a contract template to start from with your lawyer.

Ask for the template

Frequently asked questions

Who owns an AI model trained on my game?

Usually the lab. Studios typically keep ownership of the game and restrict how a model may reproduce its assets or characters.

Should I grant an AI lab exclusivity?

Only for a clear extra fee and a limited term. Without exclusivity you can license the same environment to several labs, each on an isolated copy.

Do I need my engine provider's permission?

Check your engine and middleware licences. Some terms differ for server use or modified builds, so confirm before you sign.