Skip to content
MirrorFirm

What licence questions should evaluation teams settle before using a generated release?

Licence discussions should distinguish internal evaluation, model training, contractor access and publication so intended reuse is explicit before generated material enters shared pipelines.

**Does evaluation access also cover model training?**

Do not assume it does. Ask whether the proposed licence permits internal benchmarking, supervised training, fine-tuning and other intended uses. These are questions to settle for the specific release, rather than rights implied by the fact that its content is generated.

**What about external evaluation infrastructure?**

Describe where records will be processed and who will receive them. This can include model providers, contractors and hosted evaluation services. Ask which forms of access, copying and retention are permitted, including copies embedded in debugging traces or cached prompts.

**Can results or task examples be published?**

Ask separately about aggregate findings, quoted task text, complete records and benchmark redistribution. Publication of results and redistribution of the underlying material are different uses. Derived variants and adaptations also deserve an explicit question.

A useful first step is a short intended-use inventory covering users, systems, outputs and recipients. MirrorFirm can then be asked to address that inventory in the proposed licence. Do not infer prices, duration, renewal arrangements or publication permissions from sample access or from a public content endpoint.

Map intended uses and recipients before agreeing a licence.

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

Content release: r-a220eb6e099b243b

More on Customer change handling