Jamie Njoku-Goodwin, OBE, previously served as CEO of U.K. Music and Director of Strategy to the U.K. Prime Minister at No. 10 Downing Street. Variety welcomes responsible commentary, please direct inquiries to music@variety.com.

In early 2023, I warned that failing to robustly enforce copyright could result in an epidemic of “music laundering.” As the then-Chief Executive of U.K. Music, the collective voice of the British music sector, I was fighting for the interests of creators, record labels, venues and music companies alike. Without proper guardrails, I argued, advances in artificial intelligence would enable technology companies to take people’s music without permission, use it to train their models, then generate clean, seemingly new content with its origins hidden.

At the time, I worried that the comparison with mobsters laundering stolen goods might be too dramatic. However, three years on it is becoming increasingly clear that if anything, my warning wasn’t stark enough. According to Deezer, half of all new uploads to streaming platforms are now AI generated and an extraordinary 85% of its fully AI-generated music streams are fraudulent. The music industry – and in particular artists and creators – are the ones paying the price.

Whether it is music or money, there are three stages to laundering ill-gotten gains. First comes the act of taking something illegitimately. Then it is cleaned and put through a series of processes or transactions, so no one can trace where it came from. Finally, it is integrated back into the legitimate economy and passed off as something that has been fairly acquired. AI music appears to be increasingly following this pattern.

First, in how music has been taken without permission. AI companies have persistently claimed that using copyrighted work as training data is perfectly legal. However, that argument seems to be weakening by the day. As newly unredacted court filings from the New York Times’s copyright case again OpenAI and Microsoft show, it’s not just creatives who say their work has been stolen. Even people inside those companies recognize it as such. According to the filings, in 2023 Microsoft’s own director of applied science Brent Hecht described the company’s scale of copying as “an astonishing theft of unprecedented proportions” and referred to it as “the largest theft of labor in human history.” When even the employees of tech companies are describing their actions in these terms, surely it calls time on the flimsy argument that systematically copying millions of works is simply a form of “fair use.”

When it comes to the cleaning of work, meanwhile, this month has seen a novel approach to trying to confer legitimacy on a product developed through questionable practices. Much has been made of the AI music company Suno agreeing deals with record labels to use their content. Suno, which established a dominant position in the AI music space by training its models on tens of millions of recordings without permission, even went so far as to claim that its newest version is trained only on licensed data. However, Universal and Sony allege that this version is trained using the outputs of Suno’s earlier models which, they claim, trained on their recordings. In their words, Suno has “laundered” their content.

So, what should be done?

First, it’s vital that we continue to enforce the law and not let companies use licensing deals as a form of amnesty. The message to AI companies must be clear: “take without permission first and then settle later” is not an acceptable business model.

Second is transparency. We need to know what content companies have used as training data, otherwise creators will have no ability to know when their work has been used without permission. Governments should require AI developers to disclose what their models were trained on. Banks and financial institutions have obligations on them to keep a track on where their money has come from, primarily to stop money laundering. The same principle should apply to AI companies, to combat music laundering.

Third, the industry needs to hold the line. Make no mistake, it is good that AI companies are starting to engage in licensing conversations. But those licensing deals should not absolve them of the systemic hoovering up of millions of pieces of content without permission, especially when you consider that this was how these companies managed to corner the market in the first place. According to SIQA, 93% of AI generated music comes from Suno. These companies have built a dominant market position through industrial scale copyright infringement – and even if they are finally starting to do deals, we cannot simply wipe the slate clean. Letting off AI companies now simply encourages and rewards this kind of behavior. We should all be concerned by the growing phenomenon of companies ignoring copyright law to establish a dominant market position, and then using that dominant position to set the terms for everyone else.

Ultimately, we must not lose sight of the fundamental question: in this new world of AI licensing deals, how are artists and musicians actually being paid?

In too many conversations about AI music, the creators are an afterthought. It’s vital that artists and musicians are put at the center of those conversations.

Three years ago, warnings about “music laundering” might have sounded dramatic, but today it appears to be a new form of business model. Laundering only works when people turn a blind eye to where the ill-gotten gains originally came from. Now more than ever, the music industry can’t afford to look the other way.