AI music gets labels before the law catches up

Thesis: the music industry has moved ahead of the legislator by introducing its own AI labels. That is the right direction, but a system built on self-declaration and without sanctions is easy to get around. Measurement, not the badge, will prove more durable.

The American RIAA and the international IFPI have announced they will work with Spotify and Apple Music to roll out AI markings for tracks. The proposed system is simple, and convincing for that reason. Two labels separate fully generated music from music that is merely assisted. A track that is "AI-generated", built automatically from a prompt or with a synthetic vocal and generated instrument tracks, gets a black AI icon with white capital letters. A track that is "AI-assisted", meaning mostly the work of a human, is marked by a white icon with black lowercase letters.

The mechanism works like the current explicit-content tag: artists, labels and distributors fill in the label themselves. The Recording Academy, SAG-AFTRA, the Human Artistry Campaign and A2IM have joined the coalition. The catch is that the system does not yet cover lyrics, compositions, music videos or album covers. That is the whole area where AI also works.

Voluntariness is both the strength and the weakness here. The strength: the standard can take effect in weeks rather than years, and it needs no treaties. The weakness: when a label serves the artist's interest it will appear quickly. When hiding the model's contribution serves that interest, whether through cheaper production or hype around a "human" sound, nobody has a tool to enforce it. Measurement tools fill that gap.

In June 2026 Deezer released a free tool that scans playlists on Apple Music and Spotify and flags tracks generated by AI. It is a platform that was among the first to start labelling such music itself, and it admits that other services have not followed. Sony goes further. According to Nikkei reports from February 2026, it has developed a technique for recognising which protected tracks were used in an AI song, and it can quantify the contribution, for example "30 percent Beatles and 10 percent Queen". When a developer cooperates, the technology plugs into its source system. When it does not, it compares the generated track with the existing repertoire. The goal is clear: to split revenue from AI music according to the real contribution of the original authors. Sony has not yet decided when it will start using this.

The conclusion for the market is that the dispute over AI in music will be settled not at the level of labels but at the level of measurement. A label is a declaration; measurement is proof. Whoever has the measurement sets the terms. That is why the next step should be to make such tools public and bring them under audit, so they do not turn into a black box in which the label is both judge and party.

Legislators should treat this as a test before enforcing Article 50 of the AI Act themselves. If the industry shows on its own that it can label and measure, the EU detectability requirement stops being a postulate and becomes feasible. If it does not, it will remain a set of little icons that anyone obliged to label can simply skip.