Which open music-generation models are safe for commercial scoring, and what evidence supports their training-data claims?

ACE-Step 1.5 is MIT at the repo level, DiffRhythm weights are Apache-2.0, YuE2 weights are CC BY-NC with creator-monetization permission, MusicGen weights are CC-BY-NC, Stable Audio Open uses a gated community license with CC-licensed training data. For each, what is verifiable about training-data rights and output ownership, and how would Fairly Trained-style certification or dataset audits apply? Why this is still open: Weight license texts and dataset provenance were only partly read; ACE-Step and DiffRhythm training-data provenance is unverified.