Skip to content
MirrorFirm

Can we train models on the material and publish benchmark results from it?

Permitted uses depend on the licence agreed for your engagement. Ask for training, evaluation, publication and redistribution rights to be stated explicitly in writing.

**Q: Can we train models on the material and publish benchmark results from it?**

A: That depends on the licence agreed for your engagement, and you should not assume it from a general description. Different uses can carry different terms, so ask for each to be addressed in writing.

Points to raise:

- **Evaluation only, or training as well:** these are separate uses and may be licensed separately. - **Fine-tuning and derived models:** whether models trained on the material can be deployed, shared or sold. - **Publishing results:** whether you may report scores, and whether you may show example cases or only aggregates. - **Redistribution:** whether partners, contractors or customers may receive the material, or only your own team. - **Updates and versions:** whether later releases fall under the same terms.

It also helps to be clear about what the material is. It is generated, so it is not a set of real business records, and it should not be described to others as one. It also carries no assurance of privacy or regulatory outcomes, so if you plan to publish example cases, run your own review for resemblance to real people or organisations first.

In a pilot, ask for a draft of the licence wording alongside the sample, so legal review and technical review can proceed together.

Naming your intended uses up front lets you get explicit licence wording instead of finding gaps after you have built on the material.

Audience:
Evaluation teams
Content type:
FAQ
Buyer question:
Scope and getting started

Content release: r-a220eb6e099b243b

More on Application intake