What UMG actually said
Speaking at UMG’s session at the Morgan Stanley European Technology, Media and Telecom Conference this week, EVP and chief digital officer Michael Nash framed the finding around a broader consumer expectation rather than a single feature request. “The vast majority of consumers demand transparency. They want the [AI-generated] content identified. They want the content labeled. An increasing percentage are looking for what we call a kill switch,” he said, according to Music Ally’s report of the session.
The 68% figure is UMG’s own research. It arrives days after Deezer published a survey claiming that 41% of listeners would like to “filter out 100% AI-generated music from their music streaming platform”. Both numbers describe the same desire — the ability to exclude synthetic music — but they are not measuring the same thing, and that distinction matters more than either headline.
Note what is not in the source: no methodology, no sample size, no territory breakdown, no definition of “AI-generated”. Nash’s quote is a statement of direction, not a specification.
Why 68% and 41% are not a contradiction
The temptation is to read this as UMG contradicting Deezer. A more useful reading: the two figures bracket a range, and the range itself is the signal. A kill switch is a stronger ask than a filter. “Filter out 100% AI-generated music” is a cleaning instruction; a kill switch implies a persistent, user-controlled setting that a platform honours by default. Phrase the question as a switch and expectation rises.
There is also an incentive difference worth naming. Deezer is a streaming service arguing it can act. UMG is a rights holder arguing the market demands action, which supports its negotiating position on licensing terms and platform obligations. Neither survey is neutral, and a professional reader should treat both as evidence of positioning as much as of preference.
| Source | Figure | Wording used | Who published it |
|---|---|---|---|
| UMG research | 68% | Consumers looking for a “kill switch” | UMG, via Michael Nash at the Morgan Stanley conference |
| Deezer survey | 41% | Listeners wanting to “filter out 100% AI-generated music” | Deezer, published earlier this week |
What both share is the demand for identification. On that, Nash is unambiguous: consumers want AI content identified and labelled. That is a disclosure argument, not a prohibition argument, and it is the one most likely to become policy.
What a kill switch is genuinely useful for
Strip away the polling and three practical uses emerge, each relevant well beyond the music business.
1. Licensing negotiations with a measurable ask. A rights holder can now say to a platform: labelling plus opt-out. Instead of a principle, it becomes a feature request with a survey behind it. For a label, publisher or creator licensing catalogue content, that is a cleaner clause than a vague ban — and it mirrors how brands already negotiate brand-safety controls. If you are drafting that kind of agreement, the same discipline applies as in contract clauses that protect you: define the term, define the mechanism, define who verifies it.
2. Curated listening as a product. A “human-only” or “verified human” playlist, feed or radio channel is now a plausible premium tier. Realistic example: a subscription tier promising no synthetic tracks, with a published verification method. The commercial risk is honesty — if the verification is cosmetic, the promise is a misrepresentation.
3. Provenance as a trust asset. For a creator, the kill switch debate is really a provenance debate. Being able to show where the music came from, who played what, and what tools were used becomes a differentiator rather than a compliance burden. Founders building creative tools should read this as a design requirement, not a marketing theme.
Who and what this affects
Streaming platforms face the operational question: how do you detect AI-generated audio at catalogue scale, and who bears the cost of error? A false positive on a kill switch removes a legitimate track from a paying listener; a false negative breaks the promise. Neither is cheap.
Rights holders gain a stronger hand on disclosure obligations but inherit an awkward duty: their own catalogues increasingly contain AI-assisted work. A kill switch that is too tidy will eventually catch their own releases, which is why “label it” is a safer ask than “block it”.
Artists and producers are the ones whose livelihoods sit between the two. The 41% and 68% figures describe listeners wanting control — not listeners wanting to punish artists — but the mechanism does not distinguish between an AI tool used in the mix and a fully synthetic track. Anyone whose work is partly automated should be thinking about disclosure language now.
Advisers — those working on licensing, IP and platform terms — will see this move from a survey item to a drafting item within a few contract cycles. The AI for lawyers ground is the same one: understand the mechanism before you negotiate over it.
What to watch
Confidentiality and data. Any detection system runs by analysing audio. Ask where the fingerprints and training data live, and whether an artist’s unreleased work is feeding a classifier without consent.
Accuracy. No source here gives an error rate, because none is published. Until a platform states its false-positive rate, a kill switch is a marketing claim. Treat precision as the number that decides whether this survives contact with legal challenge.
Intellectual property. Labelling obligations raise a question the survey cannot answer: what happens when an artist disputes an AI classification of their own work? That is a takedown process, and it needs a dispute route. The provenance questions here are close cousins of those in protecting creative ideas before launch.
Limits of professional use. Do not cite the 68% or the 41% in a client document as if they were established market data. They are two unpublished-methodology surveys, one from a platform and one from a major. Quote them as claims with sources, not as findings.
How and when to try it
According to Music Ally’s report, there is nothing to try yet: no product, no date, no named platform commitment. UMG has published a research position, not a roadmap. The kill switch exists as a negotiating demand and a survey result.
A realistic sequence: watch for platform announcements responding to the labelling ask, since disclosure is cheaper to ship than exclusion. If a streaming service launches a verified-human tier, test it critically — check whether the method is published, whether artists can appeal, and whether AI-assisted work is treated differently from AI-generated work.
For anyone advising on licensing or platform terms, the immediate action is unglamorous: write the definition now. “AI-generated music” needs a boundary before it becomes a clause, and the survey has not supplied one.
Sources
Frequently asked questions
What exactly did UMG announce?
UMG did not launch a product. Its EVP and chief digital officer Michael Nash said at the Morgan Stanley European Technology, Media and Telecom Conference that 68% of listeners want a kill switch for AI music, according to Music Ally. He also said consumers demand transparency, identification and labelling. It is a research claim supporting a negotiating position, not a roadmap.
How does this compare with Deezer’s survey?
Deezer claimed 41% of listeners would like to filter out AI-generated music, published earlier the same week. UMG’s figure is higher, but the questions differ: a kill switch implies a persistent setting, whereas filtering is a request. Neither source publishes methodology, sample size or territory, so the two numbers should not be read as a single market estimate.
Should creators and firms rely on these figures?
Treat them as claims, not benchmarks. Neither study publishes its method, and both publishers have commercial interests in the result. Quote them with attribution, avoid precision you cannot defend, and do not use either percentage as the basis of a client recommendation. The underlying demand for labelling is the more durable signal.
Want to apply this to your own work or your team? Tell me where you stand, and I will reply personally.
Get in touch
