This edition looks at AI music disclosure rules as it stands in November 2026, written for independent artists and small labels who run their own releases rather than handing them to a team.

A properly labelled AI record gets distributed; a hidden one eventually gets removed.

Where the money actually comes from

The central argument in music is whether a model was trained on material without permission, not whether a track was made with AI. Tools with licensed data are the safe choice.

What changed this year

Cloning a recognisable artist voice carries a different legal risk from copying a composition. Original synthetic voices are fine; digital copies of a living identity are not.

The mechanics most artists miss

More AI submissions are rejected for missing contributor credits and thin documentation than for their synthetic origin.

What to verify before release day

Stores normalise to roughly -14 LUFS on Spotify and YouTube and -16 LUFS on Apple Music, with true peak under -1.0 dBTP. AI mastering does not change that target.

How the numbers really behave

A muddy mix, a clipped take or a phase problem survives any amount of final processing. Mastering balances a finished record; it cannot finish an unfinished one.

Common mistakes to avoid

  • Keeping no record of the tool, licence and prompt used for a track.
  • Concealing AI involvement at delivery and meeting a takedown instead of a label.
  • Cloning a recognisable artist voice and describing it as a tribute.

Do this next

  1. Check the master on a phone speaker before you export it.
  2. Use only tools with a paid commercial licence and clear training data terms.
  3. Keep a provenance note per track: tool, licence, prompt, date and files.

What this means for your next release

Taken together, these are the mechanics that separate a catalogue which keeps earning from a folder of files that simply exists.