Increasingly, yes — some of it is. Through 2026 several of the largest royalty-free libraries added AI music generation next to their human catalogues. The problem is not that AI music exists. It is that most catalogues do not tell you which is which, so "royalty-free" no longer implies anything at all about who, or what, made the track.
What actually changed
For most of the last decade, a royalty-free library meant a catalogue of recordings by working composers, commissioned or licensed in, and sold on a subscription. In 2026 that stopped being a safe assumption. Artlist launched an AI music category built on Google's Lyria models, letting users generate a track from a text description and licence it under the same terms as the rest of the catalogue. Other libraries have gone a step short of that, adding AI tools that reshape, remix or extend human recordings rather than generating them from nothing.
Those two things are worth keeping separate. AI-assisted means a human recording processed by software — an automatic edit to fit your video length, say. AI-generated means no performance happened. Both get filed under the same "royalty-free" heading, and neither is reliably labelled.
How to check what you are licensing
There is no watermark and no technical test you can run. What you can do is look for a human on the other end of the track:
- Is there a named artist? Not a mood tag or a library-owned "composer collective" — a name.
- Does that name exist off the platform? A streaming profile, a back catalogue, gigs, a label, press. Generated tracks are usually credited to the library itself, to a pseudonym with no history, or to nobody.
- Does the library state a position? Some publish exactly where they stand on AI music. Silence is not the same as no.
If a track fails all three, you do not know what you have. That is not an accusation — it is just an honest description of what the metadata supports.
Three times it genuinely matters
For a lot of work — a background bed under a voiceover, a filler track in a montage — this is not worth losing sleep over. It matters in three specific situations.
Your audience or client cares. Documentary, music-adjacent content, anything with a craft audience, and increasingly any brand with a public position on AI. Being asked "who made this music?" and not being able to answer is a bad place to be.
You have a disclosure obligation. Broadcasters, agencies and public-sector clients increasingly require AI-generated material to be declared in delivery paperwork. You cannot declare what your supplier will not tell you.
The provenance could come back. The legal position of models trained on commercial recordings is unsettled and actively litigated. Nobody can tell you today how that resolves. A licence from a named artist for their own recording has none of that ambiguity — the person who made it signed the paperwork.
Where cadential. stands
There is no AI-generated music on cadential. and there will not be. Every track is a real release by a named independent artist who signed an agreement with us and takes 80% of every licence fee. You can look them up. They exist.
We are not going to pretend that is a technical advantage — a good generated track can sound fine. It is a structural one. You know who made the music, you can point at them, and the money reaches them. If that is worth something to you or your client, that is the whole pitch.
What we do not claim: we are not a Content ID clearance service, and we cannot stop a platform from flagging something. What we give you is a licence with a number on it that anyone can verify. See how that works on YouTube.