4 MIN READ
Voluntary deals were always going to fall short. The major labels, the publishers, the model makers, the independent artists, all of them have spent the last three years trying to write licences one conversation at a time, and the conversation has produced a handful of headline deals and a long tail of musicians who cannot get anyone to return their email. The statutory track is what the industry is building because the voluntary track is not finishing the job, and the regulator is the only party in the room with the standing to set a rate, define an opt out, and make the rules stick.
Here is the part worth knowing. The statutory track has three working pieces, the current proposals touch each of them, and the musicians who treat the proposals as something to act on now rather than something to wait out are the ones who will be positioned when the framework lands. The three working pieces are not the same as the three proposals. The proposals are the way the regulator is currently thinking about applying the pieces, and they are not all equally far along.
What the statutory track actually is
The statutory track runs on three mechanisms that have been in copyright law for decades, even if they were not built for generative AI. The compulsory licence gives the regulator the power to set a rate the model maker has to pay, with or without the rightsholder agreeing, on the same model as the old radio play royalty and the streaming mechanical. Collective management gives the rightsholder a single body to register with and a single body to collect from, so the working musician does not have to negotiate with every model maker individually. Rate setting is the part the Copyright Royalty Board or its equivalent handles, and it is the part that turns the compulsory licence from a concept into a number the model maker actually pays.
None of the three is new. All three were built for the previous round of disputes between rightsholders and the platforms that monetised their catalogues without asking. The reason the industry is reaching for them now is that the AI training question looks structurally similar to the streaming mechanical question from twenty years ago, and the regulator already has the legal machinery to handle it. The work is in the rate, the scope, and the opt out. The work is not in building a new agency from scratch.
What the current proposals look like
Three proposals are on the table in 2026, and they map onto the three mechanisms above. The training data opt out gives the rightsholder a way to remove a catalogue from the training set, with the model maker required to honour the opt out on the next training run, and a default that assumes training is allowed unless the rightsholder explicitly objects. The per generation royalty sets a rate the model maker pays each time a generated output is produced, regardless of whether the output is listened to, monetised, or distributed. The disclosure requirement forces the model maker to publish a list of the catalogues the model was trained on, with the rightsholder names and the percentage of training data drawn from each.
Each proposal has landed at a different stage. The opt out has the most regulator support because the legal mechanism for it already exists in the EU text and the US Copyright Office has signalled it would back a similar rule. The per generation royalty is the most contested because the rate is genuinely hard to set and the model makers are pushing back on a number that makes the economics work. The disclosure requirement has the most public support and the most industry opposition, and the disclosure fight is the one to watch in the next twelve months because it sets the precedent for the rest of the framework.
What the working musician should do
Three moves, and they are not all the same kind of move. Join the collective that represents the genre or the regional body the rightsholder works through, because the collective is who the model maker will be negotiating with once the statutory framework lands, and the rightsholder who sits outside the collective ends up on the opted in by default side of the table. Document the catalogue, every release, every alternate version, every demo, with the metadata the opt out registry will eventually require, because the documentation done now is what protects the catalogue when the opt out goes live. Opt out explicitly if the rightsholder does not want the catalogue in the training set, and do not wait for the framework to land before sending the opt out, because the first training run on the new framework will use the opt out list as of the date it is published, not the date the framework was finalised.
The three moves are not optional in the way the previous round of voluntary deals was optional. The statutory framework turns the opt out into the default, which means a musician who has not opted out is a musician whose catalogue is in the training set, and the royalty framework then treats the catalogue as licensed. The musician who has joined the collective, documented the catalogue, and sent the opt out is the musician who has a seat at the table when the rate is set. The musician who has done none of the three is the musician the statutory track will quietly pass by, and the royalty will arrive as a line item in the collective’s annual distribution rather than as a payment the musician negotiated directly.
The bottom line
Join the collective. Document the catalogue. Opt out if the rightsholder does not want to be trained on. The statutory track is going to land with or without the working musician, and the framework treats silence as consent, so the moves the musician makes now are the moves that decide whether the framework pays the musician or pays the collective on the musician’s behalf.
Sources & Further Reading
All claims in this article are sourced from primary documentation, vendor advisories, and reputable security researchers.
Spotted an error? Email the editor. Corrections are issued with a visible correction note.
Editorial standards. Every article on humanrequired.org is reviewed by a human editor before publication. AI may assist with drafting or research; final editorial control is human. Read the full standards.
