New EU rules demand transparency, but music artists still lack rights and payment
AI-generated content must be easier to trace, deepfakes must be labelled, and certain AI companies must be more transparent about their training material and methods. However, for artists, a crucial loophole remains open: the rules do not guarantee that you are asked if your music or performance is used for AI, or that you are paid for it.
By Julie Westh Nielsen
Are the voices real? Is the song made by a human? And what is the AI model that created it actually trained on? These are some of the questions that new EU regulations are set to make it easier to get answers to.
From 2 August, new transparency requirements apply to AI-generated material and thus also to musical works. For artists, there are three key changes to look out for in a new reality where AI-generated content must leave a digital watermark.
It is about clear labelling of so-called deepfakes (ed. video, image or audio recording created or manipulated using artificial intelligence), transparency about what the AI models are trained on, and that AI-generated content must carry a digital watermark.
However, the rules do not give artists and right holders the right to consent or payment for their work.
“The rules are a step forward, but they are far from solving the whole problem surrounding AI for music artists. It's good that there will be more transparency about AI. However, artists are getting neither influence nor payment with the new rules,” says Jens Skov Thomsen, Head of Secretariat and Legal Affairs at the Danish Artists' Union.
Greater transparency, but far from full transparency
The new rules are designed to make it easier to figure out when a service or producer has used AI. They set clearer requirements for transparency in cases where AI creates something that resembles reality.
If an artist's voice is artificially imitated, or a video makes it look as though an artist is doing or saying something they have never done or said, it may be a deepfake.
Danish Artist Union has previously used the term fake-art for AI content that imitates real artists.
In those cases, it must be clearly visible that the content has been created or manipulated with AI, the new rules state.
“It is of course positive that we as consumers can be clearly told when something is fake. But if a voice or performance is used without agreement, the problem isn't solved by slapping a label on it afterwards,” says Jens Skov Thomsen.
The new requirement for clear labelling does not apply to all AI-manipulated content either. For content that is not categorised as a deepfake, the ’label’ is hidden in the file's digital data. Therefore, the label is not necessarily something that you as a music consumer or rights holder can always see yourself. In other words, the new rules distinguish between AI-manipulated content and deepfakes.
”There is more transparency now, but not full transparency, and that could potentially place heavy demands on control mechanisms and the enforcement of the new rules. We have only just reached the point of figuring out whether the latest EU directive has been adequately implemented in practice,” says Jens Skov Thomsen.
Transparency is not the same as rights
EU rules also require more transparency about what certain AI models have been trained on. The companies behind them must publish a summary of their training material on their own website so that artists and other copyright holders, among others, can gain a better understanding of where the material comes from.
They must, among other things, provide information on major datasets and other central sources that the model has been trained on. Companies must also have a policy on how they comply with EU copyright rules.
However, they do not need to publish a list of every single song, film, or text they have used. As an artist, you therefore still do not necessarily get an answer to the most specific question: Has my work been used to train AI?
Danish Artist Association, together with other Danish copyright organisations, has previously demanded that AI companies provide usable knowledge about their training material, respect the artists' rights, and enter into agreements on use and payment.
Read more about DAF and DMF's reservation of rights here.
The rules apply, but who is going to keep an eye on things?
The AI Act is an EU regulation. This means that the rules already apply directly as law in Denmark. Breaches of transparency requirements, among others, can lead to fines of up to 15 million euros or three per cent of a company's global annual turnover.
However, Denmark has not yet designated the authority that is to check whether the rules on, among other things, AI labelling and deepfakes are being followed.
The Agency for Digitalisation informs the Danish Artist Union that, for the time being, the agency is responsible for providing guidance on the rules. It is the ministry that must designate the authority to carry out supervision.
The Ministry for Research, Higher Education and Digitalisation confirms that the work to designate the Danish supervisory authorities is still ongoing. According to the ministry, the previous government had introduced a bill to establish the framework for the Danish supervision, but the bill lapsed in connection with the general election. The ministry expects to reintroduce the bill as soon as possible in the coming parliamentary year.
The Danish Artist Union has also asked the ministry how the rules are specifically to be enforced in Denmark until the supervisory authority is in place. The ministry has not answered that question.
The rules therefore already apply. But Denmark has not yet put in place the control required to ensure that a central part of them is followed.
